loop-data.mjs

import { validateLoopData } from "./validate-loop-data.mjs";

export const site = {
  name: "Loop Library",
  publisher: "Forward Future",
  baseUrl: "https://signals.forwardfuture.ai/loop-library/",
  description:
    "Practical AI agent workflows for engineering, research, editorial work, evaluation, and operations.",
  updated: "2026-06-21",
  socialImageVersion: "20260621-2",
  socialImageExtension: "png",
  socialImageMimeType: "image/png",
};

export const categories = [
  { slug: "engineering", label: "Engineering" },
  { slug: "evaluation", label: "Evaluation" },
  { slug: "operations", label: "Operations" },
  { slug: "content", label: "Content" },
  { slug: "design", label: "Design" },
];

export const featuredLoopSlugs = [
  "refund-follow-up-loop",
  "five-minute-repository-maintainer-loop",
  "full-product-evaluation-loop",
];

const categorySlugByLabel = new Map([
  ["AI coding agent workflow", "engineering"],
  ["AI repository operations workflow", "engineering"],
  ["AI product evaluation workflow", "evaluation"],
  ["AI workflow design workflow", "evaluation"],
  ["AI release operations workflow", "operations"],
  ["AI data operations workflow", "operations"],
  ["AI deployment operations workflow", "operations"],
  ["AI recovery operations workflow", "operations"],
  ["AI consumer advocacy workflow", "operations"],
  ["AI search visibility workflow", "content"],
  ["AI editorial workflow", "content"],
  ["AI visual design workflow", "design"],
  ["AI frontend design workflow", "design"],
]);

export function getLoopCategory(loop) {
  const categorySlug = categorySlugByLabel.get(loop.categoryLabel);
  const category = categories.find(({ slug }) => slug === categorySlug);

  if (!category) {
    throw new Error(`No browsing category for ${loop.title}.`);
  }

  return category;
}

export const loops = [
  {
    number: "001",
    slug: "overnight-docs-sweep",
    title: "The docs sweep",
    summary:
      "Keeps documentation aligned with the current codebase and opens a reviewable pull request.",
    seoTitle: "Documentation Sweep for Coding Agents | Loop Library",
    description:
      "A reusable AI coding-agent workflow for comparing documentation with the current codebase, fixing drift, and opening a reviewable pull request.",
    categoryLabel: "AI coding agent workflow",
    author: "Matthew Berman",
    published: "2026-06-12",
    modified: "2026-06-18",
    prompt:
      "Whenever a documentation pass is needed, review the codebase in full and make sure all documentation reflects the current implementation. Update stale documentation, verify the changes, then open a pull request.",
    verifyTitle: "Documentation matches the current implementation.",
    verifyDetail: "Finish with a reviewable pull request.",
    useWhen:
      "Use this whenever implementation changes may have left READMEs, setup guides, API references, examples, or runbooks behind.",
    steps: [
      "Review implementation changes since the last documentation pass.",
      "Compare the repository's documentation with the code, configuration, commands, and behavior that now ship.",
      "Update only stale material, then verify commands, links, and examples against the current repository.",
      "Run the relevant checks and open a pull request that explains the documentation drift and the fixes.",
    ],
    why:
      "The loop ties documentation to the implementation instead of relying on memory. Requiring a pull request creates a visible diff, a review point, and a durable record of what changed.",
    note:
      "Keep the scope tied to real implementation changes. Do not rewrite accurate documentation just to create activity.",
    keywords: [
      "AI coding agent",
      "documentation audit",
      "documentation drift",
      "documentation maintenance",
      "pull request workflow",
    ],
    related: ["production-error-sweep", "architecture-satisfaction-loop"],
  },
  {
    number: "002",
    slug: "architecture-satisfaction-loop",
    title: "The architecture satisfaction loop",
    summary:
      "Refactors architecture in small, tested, independently reviewed checkpoints.",
    seoTitle:
      "Architecture Refactoring Loop for Coding Agents | Loop Library",
    description:
      "A bounded refactoring workflow that live-tests the system, runs an independent review, commits checkpoints, and records progress.",
    categoryLabel: "AI coding agent workflow",
    author: "Peter Steinberger",
    published: "2026-06-12",
    modified: "2026-06-17",
    prompt:
      "Refactor until you are happy with the architecture. After each significant step, live-test the system, run autoreview, and commit. Track progress in /tmp/refactor-{projectname}.md.",
    verifyTitle: "The architecture is satisfactory and checks pass.",
    verifyDetail:
      "Live-test, autoreview, and commit each significant step.",
    useWhen:
      "Use this for a deliberate architectural refactor where the destination can be stated in concrete terms and the current system can be tested after each meaningful change.",
    steps: [
      "Write down the architectural target, constraints, and current risks before editing code.",
      "Make one significant, reviewable change at a time.",
      "Live-test the affected behavior and run an independent review after each significant step.",
      "Commit each verified checkpoint and update the temporary progress file with decisions, blockers, and the next action.",
    ],
    why:
      "Small verified checkpoints reduce refactor risk and preserve rollback points. The progress file keeps the goal and decisions available across long sessions or handoffs.",
    note:
      "Define what satisfactory means before starting, such as module boundaries, dependency direction, passing tests, and acceptable performance. A subjective stop condition can otherwise run indefinitely.",
    keywords: [
      "AI coding agent",
      "architecture refactor",
      "autoreview",
      "incremental refactoring",
      "coding agent workflow",
    ],
    related: ["overnight-docs-sweep", "sub-50ms-page-load-loop"],
  },
  {
    number: "003",
    slug: "sub-50ms-page-load-loop",
    title: "The sub-50 ms page-load loop",
    summary:
      "Optimizes every page until it consistently loads in under 50 ms.",
    seoTitle: "Sub-50 ms Page-Load Optimization Loop | Loop Library",
    description:
      "A performance optimization workflow for coding agents that uses one repeatable benchmark and stops only when every target page meets the threshold.",
    categoryLabel: "AI coding agent workflow",
    author: "Matthew Berman",
    published: "2026-06-12",
    modified: "2026-06-17",
    prompt:
      "Continue optimizing the code for speed. After each significant change, measure page-load performance across every page under the same repeatable test conditions. Continue until every page loads in under 50 ms.",
    verifyTitle: "Every page loads in under 50 ms.",
    verifyDetail:
      "Use the same benchmark and confirm there are no regressions.",
    useWhen:
      "Use this when a product has a defined set of routes, a stable performance harness, and a 50 ms target that maps to a specific metric and environment.",
    steps: [
      "Define the exact metric, routes, test environment, warm-up behavior, and number of benchmark runs.",
      "Capture a baseline for every target page before making changes.",
      "Make one significant optimization, rerun the same benchmark, and inspect regressions across all routes.",
      "Continue until every page meets the threshold under the original test conditions.",
    ],
    why:
      "The fixed harness prevents performance work from turning into anecdotal tuning. Measuring every route after each change catches local wins that quietly slow down another page.",
    note:
      "Page load can mean server response, render completion, or a browser timing metric. Name the metric and hardware explicitly so the 50 ms target is reproducible and meaningful.",
    keywords: [
      "AI coding agent",
      "page load optimization",
      "performance benchmark",
      "web performance workflow",
      "50 ms page load",
    ],
    related: ["architecture-satisfaction-loop", "production-error-sweep"],
  },
  {
    number: "004",
    slug: "production-error-sweep",
    title: "The production error sweep",
    summary: "Finds, fixes, and verifies actionable errors in production.",
    seoTitle: "Production Error Triage Loop for Coding Agents | Loop Library",
    description:
      "A scheduled production-log workflow that traces actionable errors to root causes, verifies fixes, opens a pull request, and stops cleanly when no action is needed.",
    categoryLabel: "AI coding agent workflow",
    author: "Matthew Berman",
    published: "2026-06-12",
    modified: "2026-06-18",
    prompt:
      "Review our production logs for errors. If you find an actionable issue, trace it to its root cause, fix it, verify the fix, and open a pull request. If no actionable errors are present, stop without making changes.",
    verifyTitle: "Actionable production errors are fixed and verified.",
    verifyDetail:
      "Finish with a pull request, or stop when no actionable errors are present.",
    useWhen:
      "Use this as a scheduled reliability pass when an agent can read production telemetry, trace failures into the repository, run the relevant tests, and prepare a reviewable fix.",
    steps: [
      "Review the agreed production log window and group repeated symptoms into likely incidents.",
      "Separate actionable product errors from expected noise, transient upstream failures, and already-known issues.",
      "Trace each actionable error to a root cause, implement the smallest appropriate fix, and verify it with focused checks.",
      "Open a pull request for each verified fix. If the logs are clean, stop without making changes.",
    ],
    why:
      "The loop converts passive log review into a closed reliability workflow. It requires a root cause, verified change, and review artifact instead of stopping at a list of errors.",
    note:
      "Treat logs as sensitive production data. Do not copy credentials, tokens, personal information, or private payloads into prompts, pull requests, or chat messages.",
    keywords: [
      "AI coding agent",
      "production log review",
      "error triage",
      "root cause analysis",
      "reliability workflow",
    ],
    related: ["overnight-docs-sweep", "sub-50ms-page-load-loop"],
  },
  {
    number: "005",
    slug: "100-percent-test-coverage-loop",
    title: "The 100% test coverage loop",
    summary:
      "Adds meaningful tests until the full suite reaches 100% coverage.",
    seoTitle: "100% Test Coverage Loop for Coding Agents | Loop Library",
    description:
      "A goal-based coding-agent workflow that identifies uncovered behavior, adds meaningful tests, and stops when the full suite passes at 100% coverage.",
    categoryLabel: "AI coding agent workflow",
    author: "Matthew Berman",
    published: "2026-06-13",
    modified: "2026-06-17",
    prompt: "Add tests until we have 100% test coverage.",
    verifyTitle: "The full test suite passes at 100% coverage.",
    verifyDetail: "Use the project's coverage report as the source of truth.",
    useWhen:
      "Use this when 100% coverage is an explicit project requirement and the repository has a trustworthy coverage command, clear exclusions, and a test suite that can be run repeatedly.",
    steps: [
      "Run the complete test suite with coverage and save the baseline report.",
      "Prioritize uncovered branches and behavior by risk instead of file order.",
      "Add tests that assert meaningful outcomes, failure paths, and boundary conditions.",
      "Repeat until the full suite passes and the configured coverage report reaches 100%.",
    ],
    why:
      "A concrete coverage target gives the agent a measurable stopping condition and makes skipped code visible. Risk-first ordering keeps the work focused on behavior that matters.",
    note:
      "Coverage measures which code ran, not whether the assertions are good. Review test quality, avoid tests that only execute lines, and keep justified generated-code or platform exclusions explicit.",
    keywords: [
      "AI coding agent",
      "100 percent test coverage",
      "test coverage workflow",
      "automated testing",
      "coding agent prompt",
    ],
    related: ["architecture-satisfaction-loop", "production-error-sweep"],
  },
  {
    number: "006",
    slug: "seo-geo-visibility-loop",
    title: "The SEO/GEO visibility loop",
    summary:
      "Fixes the highest-impact gaps in search and AI answer visibility.",
    seoTitle: "SEO and GEO Visibility Audit Loop | Loop Library",
    description:
      "A repeatable search visibility workflow that fixes the highest-impact crawl, indexation, page-intent, citation, and answer-readiness gaps first.",
    categoryLabel: "AI search visibility workflow",
    author: "Matthew Berman",
    published: "2026-06-13",
    modified: "2026-06-17",
    prompt:
      "Run an SEO/GEO audit across crawlability, indexation, page intent, titles, internal links, structured data, source citations, and answer-first content. Rank the gaps by expected impact, fix the highest-leverage issue, then rerun the same crawl and target-query benchmark across search engines and AI answer engines. Repeat until no critical technical issues remain, every priority query maps to a clear answer-ready page, and the benchmark shows no high-impact gap left to fix.",
    verifyTitle:
      "Priority pages are indexable, answer-ready, and technically sound.",
    verifyDetail:
      "The repeatable crawl and query benchmark finds no remaining high-impact gaps.",
    useWhen:
      "Use this when a site has a defined set of priority pages and target questions, and you can rerun the same technical crawl and search visibility checks after each change.",
    steps: [
      "Record the target queries, answer engines, search engines, locale, date, and benchmark method.",
      "Audit crawlability, indexation, page intent, titles, internal links, structured data, citations, and visible answer quality.",
      "Rank findings by expected impact and fix one high-leverage issue at a time.",
      "Rerun the original crawl and query benchmark until no critical technical issue or high-impact content gap remains.",
    ],
    why:
      "A fixed benchmark makes visibility work measurable and prevents a long list of low-value SEO tasks from replacing the highest-impact fix. Mapping each priority query to a strong page also gives search and answer systems a clear destination.",
    note:
      "AI citations and search results vary by time, location, account state, and model. Record the test conditions and treat sampled visibility as evidence, not a guaranteed ranking.",
    keywords: [
      "SEO audit",
      "generative engine optimization",
      "GEO workflow",
      "AI search visibility",
      "answer engine optimization",
    ],
    related: ["overnight-docs-sweep", "production-error-sweep"],
  },
  {
    number: "007",
    slug: "exhaustive-logging-coverage-loop",
    title: "The logging coverage loop",
    summary: "Adds useful, tested logs to every important system path.",
    seoTitle: "Logging Coverage Loop for Coding Agents | Loop Library",
    description:
      "A goal-based observability workflow that audits important paths, adds useful structured logs, and verifies success and failure events with tests.",
    categoryLabel: "AI coding agent workflow",
    author: "Matthew Berman",
    published: "2026-06-16",
    modified: "2026-06-17",
    prompt:
      "Review the system's logging and add missing coverage until every important path produces useful, tested logs.",
    verifyTitle: "Every important path emits useful, tested logs.",
    verifyDetail:
      "Representative success and failure tests prove coverage without exposing sensitive data.",
    useWhen:
      "Use this when important user flows, service boundaries, background jobs, or failure paths are difficult to trace because the system's logging is incomplete or inconsistent.",
    steps: [
      "Inventory the important paths and define the event, outcome, severity, correlation context, and fields each one should emit.",
      "Add structured logs to uncovered paths without duplicating events or adding low-value noise.",
      "Add tests for successful and failed outcomes, then inspect representative emitted logs for useful context.",
      "Verify redaction and repeat until every important path has tested coverage or a documented reason not to log.",
    ],
    why:
      "Treating logging as testable coverage turns observability from scattered statements into a reviewable system requirement. Inspecting emitted events catches gaps that source review alone misses.",
    note:
      "Never log credentials, tokens, secrets, or sensitive personal data. Prefer stable event names and structured fields over interpolated prose.",
    keywords: [
      "AI coding agent",
      "structured logging",
      "observability coverage",
      "logging tests",
      "production diagnostics",
    ],
    related: ["production-error-sweep", "100-percent-test-coverage-loop"],
  },
  {
    number: "008",
    slug: "nightly-changelog-sweep",
    title: "The nightly changelog loop",
    summary:
      "Keeps the changelog current with meaningful changes from the previous day.",
    seoTitle: "Nightly Changelog Loop for Coding Agents | Loop Library",
    description:
      "A scheduled coding-agent workflow that reviews the previous day's changes and keeps user-facing release history complete and current.",
    categoryLabel: "AI coding agent workflow",
    author: "Matthew Berman",
    published: "2026-06-16",
    modified: "2026-06-17",
    prompt:
      "Each night, review changes from the previous day and update the changelog with anything users should know.",
    verifyTitle: "Every user-relevant change from the previous day is accounted for.",
    verifyDetail:
      "The changelog is updated and validated, or the no-change result is recorded.",
    useWhen:
      "Use this when a project changes frequently enough that user-facing release notes can drift from merged pull requests, commits, deployments, and product changes.",
    steps: [
      "Collect the previous day's merged pull requests, commits, deployments, and other in-scope changes.",
      "Identify which changes affect users and compare them with the current changelog.",
      "Add concise dated entries with useful references while preserving existing content and avoiding duplicates.",
      "Run the relevant checks and record either the validated update or the fact that no user-facing entry was needed.",
    ],
    why:
      "A daily reconciliation makes omissions visible while the context is still fresh. Limiting entries to what users should know keeps the changelog useful instead of turning it into a raw commit feed.",
    note:
      "Use the underlying change and product behavior as the source of truth. Commit titles alone can overstate, understate, or misclassify what users experienced.",
    keywords: [
      "AI coding agent",
      "nightly changelog",
      "release notes workflow",
      "changelog automation",
      "daily repository review",
    ],
    related: ["overnight-docs-sweep", "repository-cleanup-loop"],
  },
  {
    number: "009",
    slug: "quality-streak-loop",
    title: "The quality streak loop",
    summary:
      "Fixes product failures until a defined streak of realistic tests passes.",
    seoTitle: "Quality Streak Evaluation Loop for AI Products | Loop Library",
    description:
      "A realistic product-testing workflow that turns every failure into documented regression coverage and restarts the success streak after each fix.",
    categoryLabel: "AI product evaluation workflow",
    author: "Matthew Berman",
    published: "2026-06-16",
    modified: "2026-06-17",
    prompt:
      "Test realistic scenarios. When one fails, document it, add regression and benchmark coverage, fix it, and restart the streak. Stop after [N] successful cases in a row.",
    verifyTitle: "The latest [N] realistic cases pass in a row.",
    verifyDetail:
      "Every earlier failure is documented, fixed, and protected by regression and benchmark coverage.",
    useWhen:
      "Use this when product quality needs a strict consecutive-success bar and failures should permanently improve the test and benchmark suite.",
    steps: [
      "Define realistic scenarios, the quality bar, the value of [N], and the evidence required for a pass.",
      "Run cases one at a time under consistent conditions and preserve the result for review.",
      "On any failure, document it, add regression and benchmark coverage, fix the cause, verify the fix, and reset the streak to zero.",
      "Stop only after [N] consecutive cases meet the original quality bar.",
    ],
    why:
      "Restarting the streak prevents isolated successes from hiding intermittent weaknesses. Converting each failure into durable coverage makes the evaluation stronger after every miss.",
    note:
      "Choose [N] before the run and keep the scenario distribution representative. Do not lower the quality bar or avoid difficult cases to preserve the streak.",
    keywords: [
      "AI product evaluation",
      "quality streak",
      "regression testing",
      "benchmark coverage",
      "realistic scenarios",
    ],
    related: ["full-product-evaluation-loop", "100-percent-test-coverage-loop"],
  },
  {
    number: "010",
    slug: "full-product-evaluation-loop",
    title: "The full product evaluation loop",
    summary:
      "Tests every major product capability and fixes outcomes below the quality bar.",
    seoTitle: "Full Product Evaluation Loop for AI Systems | Loop Library",
    description:
      "A comprehensive product-quality workflow that evaluates realistic scenarios across every major capability, fixes weak outcomes, and reruns them to the defined bar.",
    categoryLabel: "AI product evaluation workflow",
    author: "Matthew Berman",
    published: "2026-06-16",
    modified: "2026-06-17",
    prompt:
      "Create [N] realistic scenarios covering every major capability. Before testing, define clear success criteria and choose a consistent evaluation method, such as pass/fail checks or a scoring rubric. Run every scenario under the same conditions and record evidence for each outcome. Fix the underlying cause of anything that does not meet the criteria, rerun the affected scenarios, and then rerun the complete set. Continue until every scenario meets the original quality bar.",
    verifyTitle: "Every one of the [N] scenarios meets the defined quality bar.",
    verifyDetail:
      "The final evaluated run covers every major capability under the original conditions.",
    useWhen:
      "Use this for an end-to-end product evaluation when quality must be measured across the full feature set rather than a narrow regression or a few hand-picked examples.",
    steps: [
      "List every major capability, define the success criteria and evaluation method, choose [N], and allocate realistic scenarios across the product surface.",
      "Run the full set under consistent conditions and evaluate every outcome with evidence.",
      "Document each scenario that misses the criteria, fix the underlying issue, and add focused regression coverage where appropriate.",
      "Rerun affected scenarios and then the complete set until every outcome meets the original quality bar.",
    ],
    why:
      "A fixed capability map and consistent evaluation method make product quality visible across the whole system. Requiring a final complete run catches fixes that improve one scenario while weakening another.",
    note:
      "Keep the scenario set representative and preserve failed examples. Aggregate results can hide severe misses, so require every scenario to clear the bar.",
    keywords: [
      "AI product evaluation",
      "full product testing",
      "response scoring",
      "quality benchmark",
      "feature coverage",
    ],
    related: ["quality-streak-loop", "production-data-cleanup-loop"],
  },
  {
    number: "011",
    slug: "test-suite-speed-loop",
    title: "The test-suite speed loop",
    summary:
      "Speeds up the test suite without weakening coverage, assertions, or isolation.",
    seoTitle: "Test-Suite Speed Optimization Loop | Loop Library",
    description:
      "A performance workflow for reducing test runtime under repeatable conditions without weakening coverage, assertions, isolation, or behavior.",
    categoryLabel: "AI coding agent workflow",
    author: "Matthew Berman",
    published: "2026-06-16",
    modified: "2026-06-17",
    prompt:
      "Optimize the test suite to run as quickly as possible without reducing coverage or changing behavior.",
    verifyTitle: "The suite is faster with no coverage or behavior regression.",
    verifyDetail:
      "Repeatable timing, the full passing suite, and the original coverage report prove the result.",
    useWhen:
      "Use this when slow tests are delaying local feedback or continuous integration and the project has stable commands for measuring runtime and coverage.",
    steps: [
      "Record the full-suite runtime, coverage, environment, worker settings, and repeatable timing method.",
      "Profile the suite to find expensive setup, redundant work, poor isolation, unnecessary integration paths, or safe parallelization opportunities.",
      "Make one optimization at a time, then rerun the full suite and compare timing, coverage, and behavior.",
      "Stop at the agreed runtime target or diminishing-returns rule with all original checks still passing.",
    ],
    why:
      "A fixed baseline prevents speed work from quietly trading away coverage or correctness. Profiling directs effort toward measured bottlenecks instead of speculative rewrites.",
    note:
      "Define a runtime target or diminishing-returns rule before starting. Faster tests are not an improvement if they become flaky, order-dependent, or less representative.",
    keywords: [
      "AI coding agent",
      "test suite performance",
      "faster CI",
      "test optimization",
      "coverage preservation",
    ],
    related: ["100-percent-test-coverage-loop", "sub-50ms-page-load-loop"],
  },
  {
    number: "012",
    slug: "repository-cleanup-loop",
    title: "The repository cleanup loop",
    summary:
      "Recovers valuable repository work and safely removes proven stale state.",
    seoTitle: "Repository Cleanup Loop for Coding Agents | Loop Library",
    description:
      "A repository-hygiene workflow that audits branches, pull requests, commits, and worktrees, recovers valuable changes, and removes proven stale state.",
    categoryLabel: "AI repository operations workflow",
    author: "Matthew Berman",
    published: "2026-06-16",
    modified: "2026-06-17",
    prompt:
      "Inspect local and remote branches, pull requests, commits, and worktrees. Recover valuable work and clean everything stale until the repository is current and organized.",
    verifyTitle: "Valuable work is recovered and remaining repository state is intentional.",
    verifyDetail:
      "Branches, pull requests, commits, and worktrees are current, owned, or safely removed with evidence.",
    useWhen:
      "Use this when abandoned branches, old worktrees, unclear pull requests, or unmerged commits make it difficult to know which repository state still matters.",
    steps: [
      "Inventory local and remote branches, open and recently closed pull requests, unmerged commits, and registered worktrees.",
      "Classify each item as current, valuable but unfinished, superseded, merged, abandoned, or uncertain, recording evidence and ownership.",
      "Recover valuable changes into an appropriate current branch before removing any stale reference.",
      "Clean only proven stale state, fetch and prune safely, then rerun the inventory until every remaining item is intentional.",
    ],
    why:
      "Inventory and classification separate recoverable work from clutter before cleanup begins. Repeating the inventory proves the repository is organized instead of merely smaller.",
    note:
      "Do not delete uncertain work, discard uncommitted changes, or close someone else's pull request without confirmation. Preserve evidence for every destructive cleanup action.",
    keywords: [
      "AI coding agent",
      "repository cleanup",
      "git worktree audit",
      "branch hygiene",
      "pull request triage",
    ],
    related: ["stale-safe-batch-release-loop", "nightly-changelog-sweep"],
  },
  {
    number: "013",
    slug: "stale-safe-batch-release-loop",
    title: "The stale-safe batch release loop",
    summary:
      "Batches valid changes and releases complete artifacts from the latest integrated main.",
    seoTitle: "Stale-Safe Batch Release Loop | Loop Library",
    description:
      "A release-coordination workflow that excludes stale or unfinished work, combines valid changes, and ships complete artifacts from the latest integrated main.",
    categoryLabel: "AI release operations workflow",
    author: "Matthew Berman",
    published: "2026-06-16",
    modified: "2026-06-17",
    prompt:
      "Review pending changes and pull requests, exclude stale or unfinished work, combine the valid changes, and release them together.",
    verifyTitle: "Only current, complete changes ship in the combined release.",
    verifyDetail:
      "The released revision is the latest integrated main that contains every selected change.",
    useWhen:
      "Use this when several branches or pull requests may be ready at once and the release must avoid stale worktrees, partial overlays, and incomplete changes.",
    steps: [
      "Fetch current repository and pull-request state, then inspect every candidate change for freshness, completeness, ownership, checks, and dependencies.",
      "Exclude stale, superseded, conflicting, or unfinished work and record why each candidate was omitted.",
      "Integrate the valid changes, rerun the combined checks, and select the newest main revision that contains the full batch.",
      "Release complete artifacts from a clean checkout, serialize the deployment, and verify production before closing the batch.",
    ],
    why:
      "Evaluating all candidates before integration prevents stale code from entering a release through convenience or worktree confusion. Releasing from integrated main proves the deployed artifact matches the reviewed batch.",
    note:
      "The candidate diff selects what belongs in the batch, but deployment must use complete artifacts from the latest integrated main. Never deploy from a task worktree or partial file overlay.",
    keywords: [
      "AI release operations",
      "batch release",
      "stale code prevention",
      "pull request coordination",
      "deployment safety",
    ],
    related: ["repository-cleanup-loop", "post-release-baseline-loop"],
  },
  {
    number: "014",
    slug: "production-data-cleanup-loop",
    title: "The production data cleanup loop",
    summary:
      "Removes disallowed production data and prevents the same classification errors from returning.",
    seoTitle: "Production Data Cleanup Loop for AI Systems | Loop Library",
    description:
      "A production-data quality workflow that removes disallowed records, improves classification logic, and verifies the remaining dataset against an explicit definition.",
    categoryLabel: "AI data operations workflow",
    author: "Matthew Berman",
    published: "2026-06-16",
    modified: "2026-06-17",
    prompt:
      "Review production records, remove anything that does not meet the allowed definition, improve the classification logic, and verify the remaining data.",
    verifyTitle: "Every remaining record meets the allowed definition.",
    verifyDetail:
      "Representative classification tests and a post-cleanup audit prove the retained data is valid.",
    useWhen:
      "Use this when a production dataset contains records that no longer match a product, policy, taxonomy, or quality definition and the classifier allowed them through.",
    steps: [
      "Write the allowed definition as explicit inclusion, exclusion, and edge-case rules before changing data.",
      "Audit production records, preserve a recoverable record of proposed removals, and separate clear violations from uncertain cases.",
      "Remove confirmed invalid records through the approved production path and improve the classifier with regression examples.",
      "Rerun classification tests and audit the remaining production data until every sampled and queried record meets the definition.",
    ],
    why:
      "Fixing both the existing records and the classifier closes the immediate data problem and reduces recurrence. Explicit rules and regression examples make future cleanup decisions reviewable.",
    note:
      "Follow access, retention, privacy, and audit requirements. Use backups or reversible operations where appropriate, and do not delete uncertain records without review.",
    keywords: [
      "AI data operations",
      "production data cleanup",
      "classification logic",
      "data quality audit",
      "regression examples",
    ],
    related: ["full-product-evaluation-loop", "exhaustive-logging-coverage-loop"],
  },
  {
    number: "015",
    slug: "post-release-baseline-loop",
    title: "The post-release baseline loop",
    summary:
      "Benchmarks each completed release and records a reproducible baseline.",
    seoTitle: "Post-Release Benchmark Baseline Loop | Loop Library",
    description:
      "A triggered release workflow that runs standard benchmarks against the completed release and records a reproducible baseline for future comparisons.",
    categoryLabel: "AI release operations workflow",
    author: "Matthew Berman",
    published: "2026-06-16",
    modified: "2026-06-17",
    prompt:
      "After current releases finish, run the standard benchmarks and record the results as the new baseline.",
    verifyTitle: "The new baseline belongs to the completed release.",
    verifyDetail:
      "Revision, environment, benchmark version, conditions, and results are recorded together.",
    useWhen:
      "Use this immediately after a release when future regressions or improvements need to be measured against the exact version now in production.",
    steps: [
      "Confirm every in-scope release is complete and record the production revision or artifact identity.",
      "Run the standard benchmark suite under its documented environment, data, warm-up, and repetition rules.",
      "Investigate invalid or unstable runs, then rerun only under the same documented conditions.",
      "Store the final results with the release identity and benchmark metadata, and mark them as the new comparison baseline.",
    ],
    why:
      "Tying the baseline to a verified release creates a trustworthy reference point for later performance and quality work. Recording the conditions prevents unrelated environment changes from masquerading as product changes.",
    note:
      "Do not overwrite the previous baseline until the release identity and benchmark run are verified. Keep historical baselines available for trend analysis.",
    keywords: [
      "AI release operations",
      "post-release benchmark",
      "performance baseline",
      "release verification",
      "benchmark history",
    ],
    related: ["stale-safe-batch-release-loop", "test-suite-speed-loop"],
  },
  {
    number: "016",
    slug: "ticket-to-pr-ready-loop",
    title: "The ticket-to-PR-ready loop",
    summary:
      "Turns a ticket or complaint into a verified, reviewer-ready pull request.",
    seoTitle: "Ticket-to-PR-Ready Loop for Coding Agents | Loop Library",
    description:
      "A bounded engineering workflow that turns a ticket, failing behavior, or customer complaint into a proven root cause, minimal patch, and reviewer-ready handoff.",
    categoryLabel: "AI coding agent workflow",
    author: "Hiten Shah",
    sourceUrl:
      "https://docs.google.com/document/d/1PjkOSfGaww1k_NJjswovfCdSHl31w8sxIEzXilU92gg/edit?tab=t.0",
    published: "2026-06-18",
    modified: "2026-06-19",
    prompt:
      "Take a ticket, bug report, failing behavior, or customer complaint and turn it into a review-ready patch. Reproduce the failure in the smallest representative environment, prove the root cause, make the smallest credible fix, and rerun the original reproduction plus relevant regression tests. If the issue cannot be reproduced after two serious attempts, say so. Do not fold unrelated refactors into the patch. Finish with the cause, changed files, before-and-after proof, risks, and pull-request summary.",
    verifyTitle: "The failure is fixed, verified, and ready for review.",
    verifyDetail:
      "The issue reproduces before the fix, no longer reproduces afterward, and relevant regression checks pass.",
    useWhen:
      "Use this when a real but loosely written ticket, bug report, or customer complaint needs to become a bounded engineering change with enough proof for a fast review.",
    steps: [
      "State the expected and actual behavior, then reproduce the failure in the smallest representative environment.",
      "Trace the behavior to a root cause and confirm the causal link with evidence.",
      "Implement the smallest credible fix, avoiding unrelated cleanup or hidden refactors.",
      "Repeat the original reproduction, run relevant regression checks, and package the result for review.",
    ],
    why:
      "The loop closes the gap between something being wrong and a reviewer being able to trust the patch. Reproduction, evidence, bounded scope, and a structured handoff remove the detective work from review.",
    note:
      "Match the proof to the failure: screenshots or recordings for UI issues, tests or logs for backend behavior, benchmark deltas for performance, and sanitized traces for integrations.",
    keywords: [
      "AI coding agent",
      "ticket to pull request",
      "bug reproduction",
      "root cause analysis",
      "review-ready patch",
    ],
    related: ["production-error-sweep", "quality-streak-loop"],
  },
  {
    number: "017",
    slug: "customer-ai-deployment-loop",
    title: "The customer AI deployment loop",
    summary:
      "Moves one customer AI priority through validation, controlled rollout, and monitoring.",
    seoTitle: "Customer AI Deployment Loop | Loop Library",
    description:
      "A supervised delivery workflow that advances one customer priority into a validated, gradually released AI system with monitoring, approvals, and outcome evidence.",
    categoryLabel: "AI deployment operations workflow",
    author: "AgentLed.ai Agent",
    sourceUrl:
      "https://www.agentled.ai/en/blog/post/beginners-buy-ai-automations-experts-build-ai-deployment-loops",
    published: "2026-06-18",
    modified: "2026-06-19",
    prompt:
      "Run this when a customer requests an AI workflow, reports a failure, or reaches an operations review. Choose one priority, such as enriching leads, drafting emails, summarizing meetings, or updating a CRM. Define the owner, inputs, approvals, success metric, and ROI hypothesis. Dry-run it on realistic customer data, fix the smallest verified problem, then release through approved stages and monitor production. Finish with the outcome, evidence, customer update, lessons saved, and next review.",
    verifyTitle: "One customer priority reaches a proven terminal state.",
    verifyDetail:
      "The workflow reaches its agreed rollout stage, a production issue is fixed, or a blocker is escalated with an owner and next step.",
    useWhen:
      "Use this when an AI workflow must live inside a real customer process and needs validation, approval, gradual rollout, monitoring, and a clear business outcome.",
    steps: [
      "Review the customer priority, recent feedback, workflow history, failures, approvals, usage, cost, and ROI signals.",
      "Choose one workflow or improvement and define its owner, systems, data, risk, approval gates, success criteria, and ROI hypothesis.",
      "Dry-run it on realistic customer data, repair the smallest underlying issue, and release through controlled stages.",
      "Monitor production, send the customer update, and store reusable preferences, failures, examples, and ROI observations.",
    ],
    why:
      "The workflow itself is only one part of a real deployment. This loop keeps validation, approval, rollout, monitoring, learning, and accountability tied to one customer priority.",
    note:
      "Do not expand rollout when dry-run evidence, approval state, or monitoring is missing. Keep sensitive, irreversible, financial, and customer-facing actions behind explicit human approval.",
    keywords: [
      "customer AI deployment",
      "AI workflow rollout",
      "approval gates",
      "production monitoring",
      "AI ROI",
    ],
    related: ["full-product-evaluation-loop", "quality-streak-loop"],
  },
  {
    number: "018",
    slug: "product-update-podcast-loop",
    title: "The product update podcast loop",
    summary:
      "Turns meaningful product updates into a short, source-grounded podcast episode.",
    seoTitle: "Product Update Podcast Automation Loop | Loop Library",
    description:
      "A scheduled editorial workflow that turns meaningful public product changes into a short, source-grounded podcast episode.",
    categoryLabel: "AI editorial workflow",
    author: "Pierson Marks",
    sourceUrl: "https://www.jellypod.com/mcp",
    published: "2026-06-18",
    modified: "2026-06-19",
    prompt:
      "Each night, review publicly released product changes and select only those users need to know. Verify each against the product, docs, or release notes. Use the Jellypod MCP to turn the approved changes into a three-to-five-minute podcast explaining what changed, why it matters, and how to try it. Check the script and audio for accuracy, clarity, and pronunciation. If nothing meaningful shipped, make no episode. Ask before publishing. Finish with the draft episode, sources, and review result.",
    verifyTitle: "The episode accurately covers every meaningful public update.",
    verifyDetail:
      "Finish with a review-ready three-to-five-minute episode, or a confirmed no-episode result when nothing meaningful shipped.",
    useWhen:
      "Use this when a product ships frequently enough that users would benefit from a short recurring audio explanation of what changed and how to use it.",
    steps: [
      "Collect the previous day's public product changes, documentation, and release notes.",
      "Select the changes most meaningful to users and verify what actually shipped.",
      "Use Jellypod to draft a three-to-five-minute episode covering the benefit and how to try each selected change.",
      "Review the script and audio against the sources, regenerate weak passages, and request approval before publishing.",
    ],
    why:
      "A fixed release window keeps coverage current, while editorial selection and source verification prevent the episode from becoming an automated reading of commit titles.",
    note:
      "Use only publicly released information. Do not expose private repository context, customer data, security-sensitive details, or unreleased work in the generated episode.",
    keywords: [
      "AI podcast workflow",
      "product update podcast",
      "Jellypod MCP",
      "release communication",
      "editorial automation",
    ],
    related: ["nightly-changelog-sweep", "post-release-baseline-loop"],
  },
  {
    number: "019",
    slug: "clodex-adversarial-review-loop",
    title: "The Clodex adversarial-review loop",
    summary:
      "Uses Codex to review Claude's pull request until blocking findings are resolved.",
    seoTitle: "Clodex Adversarial Code Review Loop | Loop Library",
    description:
      "A Claude-and-Codex workflow that opens a pull request, runs an independent Codex review, fixes blocking findings, and repeats.",
    categoryLabel: "AI coding agent workflow",
    author: "Lukas Kucinski",
    sourceUrl: "https://github.com/lukaskucinski/clodex",
    published: "2026-06-18",
    modified: "2026-06-19",
    prompt:
      "Run /clodex [task] think hard --max-iter 5 --threshold medium. Claude plans the task, implements it, opens a pull request, asks Codex for an adversarial review, fixes findings above the accepted severity, and repeats. Keep the branch, PR, findings, verdict, and iteration state resumable. Stop when Codex approves, only accepted findings remain, progress stalls, or the iteration cap is reached. Never describe an errored or exhausted run as approved. Finish with the PR, checks, verdict, and remaining findings.",
    verifyTitle: "The pull request reaches the configured review bar.",
    verifyDetail:
      "Codex approves it or only explicitly accepted findings remain; errors, stalls, and exhausted limits are reported as such.",
    useWhen:
      "Use Clodex when Claude is building a meaningful code change and Codex should independently review each repair round.",
    steps: [
      "Choose the task, thinking level, maximum iterations, and highest acceptable finding severity.",
      "Have Claude plan, implement, verify, and open the pull request through Clodex.",
      "Run the Codex adversarial review, fix blocking findings, push, and review again.",
      "Persist state across rounds and finish with the verdict, remaining findings, checks, and pull-request link.",
    ],
    why:
      "Clodex separates the Claude builder from the Codex reviewer and turns review feedback into a bounded repair loop. Persisted state keeps the work resumable without treating an interruption as approval.",
    note:
      "The source implementation uses Clodex with Codex as the adversarial reviewer. Treat the severity threshold as a ceiling for acceptable findings, not a minimum severity to inspect.",
    keywords: [
      "Clodex",
      "Codex adversarial review",
      "Claude Code plugin",
      "review fix loop",
      "pull request automation",
    ],
    related: ["architecture-satisfaction-loop", "stale-safe-batch-release-loop"],
  },
  {
    number: "020",
    slug: "loop-harness-verification-loop",
    title: "The Loop Harness verification loop",
    summary:
      "Ships scheduled agent work only after an independent verification pass.",
    seoTitle: "Loop Harness Second-Agent Verification Workflow | Loop Library",
    description:
      "A scheduled Loop Harness workflow that runs Claude in an isolated worktree and ships staged output only after a second Claude session verifies it.",
    categoryLabel: "AI coding agent workflow",
    author: "Istasha",
    sourceUrl: "https://github.com/lSAAGl/loop-harness",
    published: "2026-06-18",
    modified: "2026-06-19",
    prompt:
      "Use Loop Harness for scheduled repository work such as CI triage, issue grooming, dependency updates, or docs sync. Set [retry limit], then start an isolated git worktree. Let one Claude session stage a patch or outbox message and a second Claude session verify it against explicit criteria. Ship only after a pass; otherwise preserve the findings and retry only within the limit. Finish with the source revision, staged output, verifier result, delivery status, and next run.",
    verifyTitle: "Only independently verified output ships.",
    verifyDetail:
      "A second-agent pass releases the configured output; a failed verification preserves evidence and produces no external change.",
    useWhen:
      "Use this when a recurring repository task should run unattended but one agent must not be allowed to generate and approve the same output.",
    steps: [
      "Set the retry limit, wake the due Loop Harness task, and create an isolated worktree from the approved source revision.",
      "Have the primary Claude session stage one bounded result without publishing it.",
      "Have a second Claude session inspect the staged work against explicit acceptance criteria.",
      "Ship on a pass; otherwise preserve the findings, publish nothing, and retry only until the preset limit.",
    ],
    why:
      "Workspace isolation limits interference, and the second-agent gate separates generation from approval. The result can run repeatedly without relying on one session's confidence.",
    note:
      "The source implementation uses Loop Harness, git worktrees, and separate model sessions. Start with read-only tasks, test one run first, cap runtime and retries, and grant only the tools each agent needs.",
    keywords: [
      "Loop Harness",
      "scheduled coding agent",
      "git worktree isolation",
      "second-agent verification",
      "autonomous agent workflow",
    ],
    related: ["clodex-adversarial-review-loop", "overnight-docs-sweep"],
  },
  {
    number: "021",
    slug: "boeing-747-benchmark",
    title: "The Boeing 747 benchmark",
    summary:
      "Builds and improves a Three.js Boeing 747 across nine repeatable views.",
    seoTitle: "Boeing 747 Three.js Vision Benchmark | Loop Library",
    description:
      "A vision benchmark in which an agent builds a Boeing 747 from Three.js primitives, renders nine repeatable angles, and fixes what each view reveals.",
    categoryLabel: "AI visual design workflow",
    author: "@victormustar",
    sourceUrl: "https://x.com/victormustar/status/2064449741685968967",
    published: "2026-06-18",
    modified: "2026-06-19",
    prompt:
      "Before building, choose reference images, a scoring rubric, [visual threshold], and [budget]. Build the most realistic Boeing 747 you can from Three.js primitives, then create a rig that screenshots nine repeatable angles. After each change, render and score the same views, have a critic identify the weakest feature, and fix it without regressing stronger views. Keep the best version. Stop at the threshold, stalled progress, or budget. Finish with the model, nine renders, scores, remaining gaps, and run summary.",
    verifyTitle: "The Boeing 747 meets the visual bar from all nine angles.",
    verifyDetail:
      "The same camera rig and rubric show every required view meeting the preset threshold, or the run reports stagnation, budget exhaustion, and remaining gaps.",
    useWhen:
      "Use this as a concrete Three.js vision benchmark, or adapt the same capture-and-critic pattern to another rendered subject.",
    steps: [
      "Choose reference images, a scoring rubric, a visual threshold, and a budget; then build the first Boeing 747 from Three.js primitives.",
      "Create a repeatable rig that renders the same nine angles after every meaningful change.",
      "Score each view against the references, have a critic identify the weakest feature, and fix it without losing stronger work.",
      "Keep the best version and repeat until all nine views clear the visual bar or another named stop is reached.",
    ],
    why:
      "The nine-angle rig turns a subjective 3D build into a repeatable visual test. Critiquing the same views after each change exposes problems that one hero render can hide.",
    note:
      "The source run used a Boeing 747, Three.js primitives, nine camera angles, and repeated critics. To adapt it, replace the subject and renderer but keep fixed views, a visible quality bar, and preserved comparison renders.",
    keywords: [
      "Boeing 747 benchmark",
      "Three.js agent workflow",
      "vision self-verification",
      "3D reconstruction loop",
      "camera inspection system",
    ],
    related: ["quality-streak-loop", "full-product-evaluation-loop"],
  },
  {
    number: "022",
    slug: "war-loops-frontend-designer",
    title: "War Loops: frontend reconstruction",
    summary:
      "Reconstructs a real interface and repairs its weakest visual and motion mismatches.",
    seoTitle: "War Loops Frontend Reconstruction Workflow | Loop Library",
    description:
      "A War Loops workflow that captures a real page, builds a static Pencil mirror and moving Forge version, then repairs the weakest fidelity signals.",
    categoryLabel: "AI frontend design workflow",
    author: "Swayam",
    sourceUrl: "https://github.com/0xtigerclaw/war_loops",
    published: "2026-06-18",
    modified: "2026-06-19",
    prompt:
      "Point War Loops at an authorized URL or image. Capture it with a genuine browser and record the layout, styles, content, motion, and responsive behavior. Build a static Pencil mirror and a moving Forge version. Compare both with the source at desktop, tablet, and mobile sizes; repair only the weakest fidelity signals. Stop when every gate passes, progress stalls, or capture is blocked. Finish with the builds, spec, renders, scores, and remaining gaps.",
    verifyTitle: "The builds match the source across all three fidelity axes.",
    verifyDetail:
      "Static appearance, experiential motion, and responsive reflow pass their gates, or the run reports stagnation or a blocked capture.",
    useWhen:
      "Use War Loops when an authorized interface must be rebuilt from a URL or image and judged on appearance, motion, and responsive behavior.",
    steps: [
      "Capture the source with a genuine browser and extract its design spec, motion, and target viewports.",
      "Build the static Pencil mirror and moving Forge version from the verified spec.",
      "Judge both across static design, experiential motion, and responsive reflow.",
      "Repair the weakest signals without rebuilding what already matches, then repeat to a terminal fidelity decision.",
    ],
    why:
      "War Loops separates a page's still appearance from how it moves and reflows. Its surgical critic targets the weakest measured signals without churning areas that already match.",
    note:
      "The source implementation uses War Loops with Pencil and Forge. Confirm authorization to reproduce the reference, and stop on a bot wall, login gate, or unreliable capture.",
    keywords: [
      "War Loops",
      "autonomous frontend designer",
      "frontend fidelity",
      "visual evaluation loop",
      "responsive motion matching",
    ],
    related: ["full-product-evaluation-loop", "sub-50ms-page-load-loop"],
  },
  {
    number: "023",
    slug: "self-improving-champion-loop",
    title: "The self-improving champion loop",
    summary:
      "Promotes prompt or policy changes only when they win on fresh holdout cases.",
    seoTitle: "Self-Improving Champion Evaluation Loop | Loop Library",
    description:
      "A prompt-optimization workflow that tests challengers on a working set, promotes only fresh holdout wins, and keeps the current champion on uncertainty.",
    categoryLabel: "AI product evaluation workflow",
    author: "Jose C. Munoz",
    published: "2026-06-18",
    modified: "2026-06-19",
    prompt:
      "Improve a prompt, policy, or configuration. A support assistant's system prompt is one example. Save the champion, its score, a working set, untouched holdout cases, must-pass checks, and [budget]. Each round, change one thing based on a recorded failure. Promote the challenger only if it beats the champion on holdouts by [margin] without weakening a must-pass check; otherwise keep the champion. Stop at the target, budget limit, or no progress. Return the winner, scores, experiment log, and remaining failures.",
    verifyTitle: "The best holdout-tested champion is returned.",
    verifyDetail:
      "Every challenger is logged, and accepted changes beat the previous champion on untouched cases without weakening a must-pass check.",
    useWhen:
      "Use this to tune a prompt, policy, or configuration when cheap iteration is useful but final acceptance must use fresh examples.",
    steps: [
      "Save the current champion, working set, untouched holdout cases, must-pass checks, improvement margin, budget, and experiment log.",
      "Use a recorded failure to propose one targeted challenger and test it on the working set.",
      "Freeze promising challengers and evaluate them on the untouched holdout cases and every must-pass check.",
      "Promote only a meaningful, regression-free holdout win; log every result and return the champion at the stop condition.",
    ],
    why:
      "Separating the working set from fresh holdout cases limits overfitting. Keeping the current best by default prevents regressions, while a fixed budget bounds the search.",
    note:
      "Keep the working set and holdout cases separate: edit against the former, judge final acceptance on the latter. Choose the budget and margin before starting, and do not weaken a must-pass check after a failed challenger.",
    keywords: [
      "self-improving loop",
      "champion challenger evaluation",
      "Goodhart prevention",
      "independent evaluation gate",
      "bounded optimization workflow",
    ],
    related: ["full-product-evaluation-loop", "quality-streak-loop"],
  },
  {
    number: "024",
    slug: "devils-advocate-design-loop",
    title: "The devil's-advocate loop",
    summary:
      "Challenges a design until every high-impact objection is resolved or explicitly accepted.",
    seoTitle: "Devil's-Advocate Design Review Loop | Loop Library",
    description:
      "A critic-and-builder workflow that attacks a design, tracks every objection, and requires evidence before an objection can be closed.",
    categoryLabel: "AI product evaluation workflow",
    author: "Anonymous contributor",
    published: "2026-06-18",
    modified: "2026-06-19",
    prompt:
      "Before committing to an architecture, interface, or rollout plan, have a critic argue that it is wrong. Record each objection, impact, and status in a repository-local log at .agent-reviews/redteam.md. The builder must fix and verify each high-impact weakness or document why it is accepted; the critic may reopen unsupported answers. Stop when no high-impact objection remains or the same issues repeat for two rounds without new evidence. Finish with the decision, resolved and accepted objections, evidence, and any stalemate.",
    verifyTitle: "No high-impact objection remains open.",
    verifyDetail:
      "Every logged objection is verified as resolved or explicitly accepted with evidence, or the final report truthfully records a two-round stalemate.",
    useWhen:
      "Use this before committing to an architecture, interface, rollout plan, or other consequential design that benefits from structured adversarial review.",
    steps: [
      "Write the design goals and acceptance criteria, then initialize .agent-reviews/redteam.md inside the repository and keep it out of commits.",
      "Have the critic present the strongest evidence-backed case against the current design and rank each objection by impact.",
      "Have the builder repair the weakness or document an explicit acceptance rationale, then verify the result against the stated criteria.",
      "Let the critic reopen weak answers and repeat until the objections are closed with evidence or the loop reports a stalemate honestly.",
    ],
    why:
      "Separating critic and builder roles makes disagreement explicit. A persistent objection log prevents circular debate, while evidence-based closure stops the builder from declaring success by explanation alone.",
    note:
      "Keep the critic independent where possible. Do not change the acceptance criteria mid-run simply to close a difficult objection.",
    keywords: [
      "devil's advocate loop",
      "adversarial design review",
      "critic builder workflow",
      "architecture objection log",
      "red team design process",
    ],
    related: ["architecture-satisfaction-loop", "clodex-adversarial-review-loop"],
  },
  {
    number: "025",
    slug: "fresh-clone-loop",
    title: "The fresh-clone loop",
    summary:
      "Repeats clean onboarding from the README until no hidden setup assumptions remain.",
    seoTitle: "Fresh Clone README Verification Loop | Loop Library",
    description:
      "A disposable-environment workflow that follows the README from scratch, fixes every hidden setup assumption, and restarts until onboarding works cleanly.",
    categoryLabel: "AI repository operations workflow",
    author: "0xUmbra",
    published: "2026-06-18",
    modified: "2026-06-19",
    prompt:
      "Clone [repository] into a disposable environment and follow only its README to the documented ready state, such as running the app or building the package. When a step fails or assumes missing knowledge, record the gap, fix the setup or documentation issue, discard the environment, and start again. Carry no dependencies, configuration, credentials, or repairs between attempts. Stop when one uninterrupted fresh clone reaches that state, progress stalls, or [budget] ends. Return exact commands, gaps closed, and remaining blockers.",
    verifyTitle: "A clean environment reaches the documented ready state using only the README.",
    verifyDetail:
      "The final run uses only the onboarding guide and needs no unstated dependency, configuration, or manual repair.",
    useWhen:
      "Use this to test whether a repository's onboarding instructions work in a clean environment without undocumented help.",
    steps: [
      "Create a disposable environment with no project dependencies or configuration carried over from another checkout.",
      "Fresh-clone the repository and follow only the README, recording every missing step, hidden assumption, and failure.",
      "Fix the smallest setup or documentation gap, discard the environment completely, and begin again.",
      "Repeat until one clean run reaches the documented ready state without intervention, then report the exact commands and gaps closed.",
    ],
    why:
      "Destroying the environment after each repair prevents local state from hiding the next problem. The final uninterrupted run is direct evidence that the README, not the operator's memory, is sufficient.",
    note:
      "Use an isolated disposable environment and review the repository before executing it. Never copy personal credentials into the test environment or run untrusted setup scripts on a production host.",
    keywords: [
      "fresh clone loop",
      "README verification",
      "developer onboarding test",
      "clean environment setup",
      "repository documentation workflow",
    ],
    related: ["overnight-docs-sweep", "repository-cleanup-loop"],
  },
  {
    number: "026",
    slug: "infinite-clickbait-loop",
    title: "The Infinite Clickbait thumbnail loop",
    summary:
      "Iterates thumbnail concepts until one clears the quality bar without misleading viewers.",
    seoTitle: "Infinite Clickbait Thumbnail Iteration Loop | Loop Library",
    description:
      "A thumbnail workflow that creates ten concepts, scores the top three against a relevant YouTube channel, and improves the winner without misleading viewers.",
    categoryLabel: "AI visual design workflow",
    author: "@Alex_FF",
    published: "2026-06-18",
    modified: "2026-06-19",
    prompt:
      "For [video], use [approved assets] to make ten thumbnail concepts. Score each at real YouTube sizes against [inspiration channel] for clarity, curiosity, emotional pull, contrast, and accuracy. Take the top three, improve each one's weakest dimension, and rescore them under the same rubric. Keep iterating the strongest concept until it clears [quality threshold] or [budget] ends. Reject anything the video cannot deliver. Return the winner, two runners-up, previews, final scores, and rationale.",
    verifyTitle: "One accurate thumbnail clears the fixed quality threshold.",
    verifyDetail:
      "The winner outscores the alternatives under the same conditions, remains legible at realistic sizes, and represents the video accurately.",
    useWhen:
      "Use this when a video topic and asset set are ready but the thumbnail needs several structured ideation and critique rounds before production.",
    steps: [
      "Define the video subject, approved assets, inspiration channel, quality threshold, budget, and five-part rubric.",
      "Create ten distinct concepts, inspect them at real YouTube sizes, and score each one under the same conditions.",
      "Select the top three, improve the weakest dimension of each, and rescore them.",
      "Stop at the quality bar or budget, reject misleading concepts, and return the winner plus two runners-up.",
    ],
    why:
      "A varied first set creates real options, while a fixed rubric makes later rounds comparable. Scoring accuracy prevents curiosity from becoming a promise the video cannot keep.",
    note:
      "Choose an inspiration channel whose audience and visual language are relevant. Evaluate the actual thumbnail crop at desktop and mobile sizes, and reject concepts that misrepresent the video's substance.",
    keywords: [
      "Infinite Clickbait",
      "YouTube thumbnail loop",
      "thumbnail iteration workflow",
      "clickbait scoring rubric",
      "AI visual design",
    ],
    related: ["boeing-747-benchmark", "full-product-evaluation-loop"],
  },
  {
    number: "027",
    slug: "autonomy-loop",
    title: "The autonomy-loop builder-reviewer loop",
    summary:
      "Passes code between builder and reviewer until tests prove each accepted fix.",
    seoTitle: "autonomy-loop Builder-Reviewer Workflow | Loop Library",
    description:
      "An autonomy-loop workflow in which a builder and adversarial reviewer pass a git baton between worktrees and prove each new test can catch its fix.",
    categoryLabel: "AI coding agent workflow",
    author: "@inferencegod",
    sourceUrl: "https://github.com/inferencegod/autonomy-loop",
    published: "2026-06-19",
    modified: "2026-06-19",
    prompt:
      "Use autonomy-loop for [repository task] after the test, build, and lint gates pass. Run /autonomy-loop:autonomy-init, then start builder and reviewer in separate worktrees. The builder reads LOOP-STATE.md, makes one bounded change, and adds a red-before, green-after test. The reviewer reruns the gates and proves the test by reverting or mutating the fix. Accept only on both passes; park protected or repeated-failure work for a human. Finish with the commit, gate evidence, test proof, trust tier, and risks.",
    verifyTitle: "Every accepted wave passes autonomy-loop's proof-of-test gate.",
    verifyDetail:
      "The new test fails without the change, passes with it, every configured gate passes, and protected production changes remain human-gated.",
    useWhen:
      "Use autonomy-loop when a repository has deterministic test, build, and lint gates plus a task suited to repeated builder-reviewer handoffs.",
    steps: [
      "Initialize autonomy-loop, configure deterministic gates and protected paths, and create separate builder and reviewer worktrees.",
      "Have the builder read LOOP-STATE.md, implement one bounded change, add a red-before, green-after test, and hand off.",
      "Have the reviewer rerun every gate and use revert-or-mutate proof to show the test catches the change.",
      "Accept only on both passes; otherwise return findings or park the wave for a human when a circuit breaker fires.",
    ],
    why:
      "Separate worktrees and a git-backed LOOP-STATE.md baton keep the roles independent and resumable. The revert-or-mutate check catches tests that execute code without proving the fix.",
    note:
      "The source implementation uses autonomy-loop commands, separate worktrees, and a git-backed baton. Treat local hooks as tripwires, not a security boundary, and keep protected changes behind enforced approval.",
    keywords: [
      "autonomy-loop",
      "adversarial code review",
      "mutation testing",
      "builder reviewer workflow",
      "Claude Code loop",
    ],
    related: ["clodex-adversarial-review-loop", "loop-harness-verification-loop"],
  },
  {
    number: "028",
    slug: "codex-completion-contract-loop",
    title: "The Codex completion-contract loop",
    summary:
      "Defines completion up front and requires evidence for every reported result.",
    seoTitle: "Codex Completion Contract and Evidence Loop | Loop Library",
    description:
      "A goal-planner-codex workflow that defines completion up front, tracks proof for every requirement, and prevents partial Codex work from being reported as done.",
    categoryLabel: "AI coding agent workflow",
    author: "3goblack (@Dis_Trackted)",
    sourceUrl:
      "https://github.com/ranvier2d2/skills-share/tree/main/skills/goal-planner-codex",
    published: "2026-06-19",
    modified: "2026-06-19",
    prompt:
      "Run $goal-planner-codex [task] for long-running Codex work where partial work could be mistaken for done. Landing a PR and verifying production is one example. Before acting, define every required outcome and its evidence. After each bounded action, mark requirements proved, weak, missing, or contradicted. Complete the Goal only when all are proved; otherwise stop as blocked, stalled, or exhausted. Ask before creating Goal state. Finish with the requirement-to-evidence table, status, owner, and next action.",
    verifyTitle: "Every Codex Goal requirement has current, adequate proof.",
    verifyDetail:
      "The final audit contains no weak, missing, or contradicted required item; otherwise the work remains open, blocked, or exhausted.",
    useWhen:
      "Use this for long-running Codex work, pull requests, runtime checks, or user-visible artifacts where a plausible partial result could be mistaken for completion.",
    steps: [
      "Recover a measurable definition of done for every ambiguous requirement.",
      "Record the requirements, scope, non-goals, evidence plan, and current status without expanding the requested work.",
      "Execute one bounded action at a time and attach current evidence to each affected requirement.",
      "Audit every requirement before closure and preserve honest blocked, exhausted, stalled, or contradicted states.",
    ],
    why:
      "A durable completion contract keeps the definition of done visible across long sessions. Mapping every requirement to evidence makes false completion easy to detect.",
    note:
      "Use $goal-planner-codex only when the user explicitly asks for a Codex Goal or completion audit. Create native Goal state only with approval; ordinary task planning does not need it, and budget exhaustion never counts as success.",
    keywords: [
      "Codex Goal",
      "completion contract",
      "evidence audit",
      "definition of done",
      "false completion prevention",
    ],
    related: ["ticket-to-pr-ready-loop", "quality-streak-loop"],
  },
  {
    number: "029",
    slug: "revolve-self-improvement-loop",
    title: "The Revolve versioned-experiment loop",
    summary:
      "Improves prompts, code, or configurations through comparable, checkpointed experiments.",
    seoTitle: "Revolve Versioned Experiment Loop | Loop Library",
    description:
      "A Revolve workflow that improves prompts, code, or configurations through checkpointed experiments whose scores remain comparable across sessions.",
    categoryLabel: "AI product evaluation workflow",
    author: "Agent Zero",
    sourceUrl: "https://github.com/agent0ai/revolve",
    published: "2026-06-19",
    modified: "2026-06-19",
    prompt:
      "Use Revolve to improve a support prompt, code path, or testable subject. In revolve/, define the goal and [budget], freeze the tests and scoring, checkpoint the current version, and record a baseline. Each round, test one hypothesis; keep only a clear, regression-free win. If the evaluation changes, open a new revision and rerun the baseline. Ask before changing live files. Stop on success, no progress, a blocker, or exhausted budget. Return the best checkpoint, comparisons, rollback, and next action.",
    verifyTitle: "The best Revolve checkpoint wins within one evaluation revision.",
    verifyDetail:
      "The incumbent and candidates have comparable recorded runs, accepted changes pass every guard, rollback is available, and live promotion has approval.",
    useWhen:
      "Use Revolve to improve a prompt, policy, workflow, model configuration, code path, or dataset when experiments must remain comparable and resumable across sessions.",
    steps: [
      "Create or resume revolve/, define the objective and permissions, freeze an evaluation revision, checkpoint the incumbent, and record its baseline.",
      "Choose one evidence-backed hypothesis, create a candidate checkpoint, and test it under the unchanged revision.",
      "Promote internally only on a meaningful guard-safe win; if the evaluation changes, open a new revision and rerun the incumbent.",
      "Stop on a named condition, and require explicit approval plus verification before changing live files.",
    ],
    why:
      "Revolve's revision boundaries prevent scores from different tests or rubrics from being compared as equivalent. Checkpoints and an internal-before-live promotion boundary keep long-running research resumable and reversible.",
    note:
      "The source examples include improving CLI error messages, reducing image-export latency, tuning a support-assistant prompt, and hardening a parser. Replace the subject and metric, but keep the revision, checkpoint, and rollback discipline.",
    keywords: [
      "Revolve",
      "agent self improvement",
      "checkpoint evaluation",
      "revisioned experiments",
      "evidence based promotion",
    ],
    related: ["self-improving-champion-loop", "full-product-evaluation-loop"],
  },
  {
    number: "030",
    slug: "five-minute-repository-maintainer-loop",
    title: "The five-minute repository maintainer loop",
    summary:
      "Keeps repository work moving through dedicated threads without interrupting active agents.",
    seoTitle: "Five-Minute Repository Maintainer Loop | Loop Library",
    description:
      "A five-minute Codex workflow that triages repositories, directs bounded maintenance to dedicated threads, and requires proof and permission before work lands.",
    categoryLabel: "AI repository operations workflow",
    author: "Peter Steinberger",
    sourceUrl:
      "https://github.com/steipete/agent-scripts/blob/main/skills/maintainer-orchestrator/SKILL.md",
    published: "2026-06-19",
    modified: "2026-06-19",
    prompt:
      "While repository maintenance is active, wake every five minutes. Triage [repositories] and read each repository thread's latest state. Reuse one thread per repository; assign its highest-value bounded task only within granted permissions, and do not interrupt coherent active work. Require tests, live proof, autoreview, and green CI before work can land. Escalate product, access, security, or irreversible decisions. Record meaningful changes and stop when every item is landed, decision-ready, blocked, or has no work.",
    verifyTitle:
      "Every repository item reaches a proven handoff or terminal state.",
    verifyDetail:
      "Authorized autonomous work lands with evidence; other items are decision-ready, blocked with one exact ask, or recorded as a clean no-op.",
    useWhen:
      "Use this when Codex may coordinate maintenance across several active repositories and you want parallel work to stay steerable without duplicating or micromanaging threads.",
    steps: [
      "Define the repository scope, exclusions, and separate permissions for triage, delegation, implementation, push, CI repair, merge, and release.",
      "Every five minutes, refresh each repository queue and read the latest state of its existing thread before choosing the highest-value eligible item.",
      "Reuse one thread per repository, assign one bounded task, and let coherent active work continue unless it is blocked, stalled, unsafe, or off course.",
      "Require tests, live proof, autoreview, and green CI; record the evidence, then route the next item or present the owner with one exact decision.",
    ],
    why:
      "A five-minute heartbeat keeps the control plane current without turning polling into micromanagement. One thread per repository preserves context, while proof and authorization gates make autonomous landing auditable.",
    note:
      "The source pairs Maintainer Orchestrator with github-project-triage, autoreview, and computer use for live proof. A heartbeat automates observation, not authority: triage, delegation, implementation, push, merge, and release remain separate permissions. Read current thread state before steering, and never duplicate or interrupt active work.",
    keywords: [
      "Codex repository maintenance",
      "multi-repository orchestration",
      "five minute agent loop",
      "GitHub project triage",
      "thread delegation",
    ],
    related: ["ticket-to-pr-ready-loop", "stale-safe-batch-release-loop"],
  },
  {
    number: "031",
    slug: "recent-feedback-sweep",
    title: "The recent-feedback sweep",
    summary:
      "Turns recent user corrections into a project-wide audit and verified fixes.",
    seoTitle: "Recent-Feedback Project Audit | Loop Library",
    description:
      "A project audit that turns recent user-reported problems into reusable failure patterns, fixes every confirmed match, and verifies a clean final sweep.",
    categoryLabel: "AI coding agent workflow",
    author: "Matthew Berman",
    published: "2026-06-19",
    modified: "2026-06-19",
    prompt:
      "Review all available threads from [lookback window] where I reported something wrong with [project] and asked for a fix. Build a deduplicated issue list, group it into failure patterns, and verify current state. Audit the complete project for every pattern, fix each confirmed instance, and add regression coverage where practical. Repeat the full audit until it finds no remaining instance or [iteration budget] ends. Stop on blocked or approval-gated work. Return the issues, fixes, evidence, and blockers.",
    verifyTitle:
      "The issue inventory is closed and a fresh pattern audit is clean.",
    verifyDetail:
      "Every reported issue and newly found match has current proof of resolution; blocked, approval-gated, or budget-exhausted items remain explicitly open.",
    useWhen:
      "Use this after several days of project feedback when repeated mistakes may point to similar issues elsewhere and the agent can inspect both the conversation history and the complete current project.",
    steps: [
      "Define the lookback window and complete project surface, then collect every accessible thread in which the user reported a problem and requested a fix.",
      "Deduplicate the reported issues, verify their current status, and turn the concrete examples into explicit failure patterns and audit checks.",
      "Audit every in-scope project surface for each pattern, fix one confirmed instance at a time, and add regression coverage where practical.",
      "Run targeted checks after each fix, then rerun the complete pattern audit and relevant full checks before declaring the sweep clean.",
    ],
    why:
      "Recent corrections are concrete examples of the quality bar the project missed. Grouping them into failure patterns turns one-off feedback into a reusable audit rubric, while a fresh full sweep catches sibling defects and verifies the current project rather than trusting old thread state.",
    note:
      "Thread access and a complete surface inventory are prerequisites. Do not infer defects from neutral discussion, reopen resolved issues without checking current behavior, or claim success while an inaccessible, blocked, approval-gated, or budget-exhausted item remains. Get approval before destructive, production, or external actions.",
    keywords: [
      "recent user feedback",
      "project-wide issue audit",
      "failure pattern sweep",
      "regression prevention",
      "AI coding agent",
    ],
    related: ["full-product-evaluation-loop", "quality-streak-loop"],
  },
  {
    number: "032",
    slug: "promise-to-proof-loop",
    title: "The promise-to-proof loop",
    summary:
      "Checks whether every customer-facing claim is true, then fixes the riskiest mismatch first.",
    seoTitle: "Promise-to-Proof Product Audit | Loop Library",
    description:
      "A product review that compares claims in marketing, documentation, demos, and AI answers with current evidence, then fixes or narrows unsupported promises.",
    categoryLabel: "AI product evaluation workflow",
    author: "Felix Haeberle (@felixhaberle)",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "List every customer-facing promise [product] makes in marketing, documentation, demos, and AI answers. Compare each promise with current product behavior and evidence, then label it proven, partly proven, misleading, unsupported, outdated, or missing evidence. Fix or narrow the riskiest mismatch and rerun the affected check. Repeat until no high-risk unsupported promise remains. Ask before changing production or public copy. Return the promises, evidence, fixes, and decisions needed.",
    verifyTitle:
      "Every high-risk customer promise is supported, narrowed, or waiting on an explicit decision.",
    verifyDetail:
      "Each promise links to current evidence, and every high-risk mismatch is fixed, narrowed to what the product can prove, or clearly approval-gated.",
    useWhen:
      "Use this when what a product says it does may no longer match what it actually does across marketing, documentation, demos, support answers, or the live product.",
    steps: [
      "List the promises customers can see and rewrite each one as a concrete expectation, such as a feature working, a limit being honored, or an answer being accurate.",
      "Compare each expectation with current product behavior, code, tests, documentation, examples, logs, or other direct evidence; do not guess.",
      "Rank mismatches by the harm they could do to customer trust, then fix the riskiest one or narrow the public promise to what the product can prove.",
      "Rerun the same check and repeat until no high-risk unsupported promise remains, progress is blocked, or the next action needs approval.",
    ],
    why:
      "This turns a vague question—can customers trust what we say?—into a list of promises that can each be checked. Fixing one risky mismatch at a time keeps the product and its public explanation aligned without turning the audit into an uncontrolled rewrite.",
    note:
      "Evidence can include live product behavior, tests, documentation, logs, screenshots, or reproducible examples. A promise may be supported, narrowed, or removed; the product does not always need to change. Production changes and public publication still require approval.",
    keywords: [
      "product promise audit",
      "customer trust",
      "claim verification",
      "evidence based product review",
      "marketing product alignment",
    ],
    related: ["full-product-evaluation-loop", "recent-feedback-sweep"],
  },
  {
    number: "033",
    slug: "propagation-compliance-loop",
    title: "The propagation compliance loop",
    summary:
      "After one value changes, finds every other place that still shows the old value.",
    seoTitle: "Repository Propagation Compliance Loop | Loop Library",
    description:
      "A consistency check for values copied across a code project: update every affected copy, find leftovers, and prove that only intentional old references remain.",
    categoryLabel: "AI coding agent workflow",
    author: "@iamTristan",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "After changing a version, count, rule, name, or configuration, list where the new value belongs and update it. Search the project for the old value and related forms. Review each match: fix real stale values, but keep intentional history, examples, migrations, or compatibility rules. Repeat until zero stale values remain. If one returns for two rounds, stop and identify what may be regenerating it. Return changes, intentional matches, and search output.",
    verifyTitle: "No unintended copy of the old value remains.",
    verifyDetail:
      "The final searches find only references that are intentionally historical or required for examples, migrations, or compatibility, with a reason recorded for each one.",
    useWhen:
      "Use this after changing something that appears in several files—such as a version number, feature name, count, rule, setting, or identifier—and every copy must stay consistent.",
    steps: [
      "List the files, documentation, settings, generated outputs, or operational notes that are expected to copy the changed value.",
      "Update the known copies, then search the whole project for the old value, old spelling, and other likely leftover forms.",
      "Decide whether each match is truly stale or intentionally preserved history, an example, a migration, or a compatibility rule; fix only the stale matches.",
      "Repeat the same searches until no stale match remains; if one comes back for two rounds, stop and identify the generator or process restoring it.",
    ],
    why:
      "The repeat search is the important part: it catches copies missed by the first update. Reviewing each match also prevents a broad replacement from corrupting historical notes, migration code, or examples that intentionally show the old value.",
    note:
      "The exact files depend on the change. The original submission used several operational notes and procedure files; another project might need code, tests, documentation, deployment settings, generated files, or all of them.",
    keywords: [
      "configuration propagation",
      "version update audit",
      "stale value search",
      "repository consistency",
      "grep verification loop",
    ],
    related: ["overnight-docs-sweep", "recent-feedback-sweep"],
  },
  {
    number: "034",
    slug: "multi-llm-convergence-loop",
    title: "The multi-LLM convergence loop",
    summary:
      "Has two different AI systems review the same work until both approve one unchanged version.",
    seoTitle: "Multi-LLM Convergence Review Loop | Loop Library",
    description:
      "Alternates two AI systems from different providers to review a plan, document, or code change until both approve the exact same version.",
    categoryLabel: "AI product evaluation workflow",
    author: "Donn Felker (@donnfelker)",
    sourceUrl: "https://github.com/donnfelker/loop-skills",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "Review [plan, specification, document, or code change] against [quality bar] for at most [pass limit] rounds. Have one of two genuinely different model families—AI systems from separate providers—review it. Verify each finding and apply only necessary fixes, then give the revised version to the other reviewer. Succeed only when both approve the same unchanged version. Stop at the limit, repeating disagreement (oscillation), unavailable review, or required approval. Return the final work, round log, verdict, and disagreements.",
    verifyTitle:
      "Two different AI model families approve the exact same version.",
    verifyDetail:
      "The final two clean reviews come from different model families with no edit between them; a pass limit, repeating disagreement, unavailable reviewer, or approval boundary is reported as a stall instead of consensus.",
    useWhen:
      "Use this when an important plan, specification, design, document, or code change benefits from two independent AI perspectives rather than one model reviewing its own blind spots.",
    steps: [
      "Choose the work being reviewed, define what counts as acceptable, set a maximum number of rounds, and gather the source material reviewers should trust.",
      "Give the current version to the first AI model family, check whether each finding is valid, apply only necessary fixes, and record the round.",
      "Give the resulting version to the other model family; if either reviewer causes another edit, both must review the new version again.",
      "Finish only when both independently approve one unchanged version; otherwise stop at the round limit, repeated back-and-forth, reviewer failure, or an approval boundary.",
    ],
    why:
      "Different model families can notice different problems. Requiring both to approve the exact same version prevents a clean review of an older draft from being counted as approval of a newer one, and the round log shows how the agreement was reached.",
    note:
      "A model family means a genuinely separate model lineage, such as a Codex/OpenAI reviewer and a Claude/Anthropic reviewer—not two prompts sent to the same underlying model. With only one family, label the result a single-model review and do not claim consensus.",
    keywords: [
      "multi LLM review",
      "cross model consensus",
      "artifact convergence",
      "alternating reviewers",
      "independent AI review",
    ],
    related: ["clodex-adversarial-review-loop", "devils-advocate-design-loop"],
  },
  {
    number: "035",
    slug: "goal-forge-loop",
    title: "The Goal Forge loop",
    summary:
      "Turns a rough coding idea into measurable planning files before Codex starts a long run.",
    seoTitle: "Goal Forge Specification Loop for Codex | Loop Library",
    description:
      "A planning workflow that interviews the user, writes what should be built in SPEC.md, and writes how Codex should execute and verify it in GOAL.md.",
    categoryLabel: "AI coding agent workflow",
    author: "michael Guo (@michaelzsguo)",
    sourceUrl: "https://github.com/michaelpersonal/goal-forge",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "Turn [rough coding idea] into two planning files before Codex starts /goal, its long-running task mode. Interview the user, then write SPEC.md: what to build, exclude, and consider, plus measurable done_when completion checks. Write GOAL.md: the work plan, progress scorecard, quick and final checks, memory files, evidence, and approval boundaries. If any key decision, permission, tool, environment requirement, or test is missing, stop as not ready. Do not start implementation without approval.",
    verifyTitle:
      "The planning files say what to build, how to judge it, and when to stop.",
    verifyDetail:
      "Every done_when completion check names observable evidence, the quick and final checks can actually run, the environment is ready, and unresolved decisions are clearly marked not ready.",
    useWhen:
      "Use this when a rough coding idea is too vague to hand to Codex for a long autonomous run and the user first needs to settle scope, completion checks, safety boundaries, and required tools.",
    steps: [
      "Ask the user what the finished feature should do, what is out of scope, which edge cases matter, what could go wrong, and what evidence would prove completion; write those decisions in SPEC.md.",
      "Point out ambiguous requirements with concrete interpretations and have the user resolve product decisions instead of letting the coding agent silently choose.",
      "Write GOAL.md with the ordered work, a progress scorecard, quick checks for each iteration, slower final checks, memory files for long runs, approval boundaries, and required evidence.",
      "Confirm that the tools, permissions, environment, and tests exist; stop as not ready when anything essential is missing, and start the long-running task only after approval.",
    ],
    why:
      "Goal Forge makes the user decide what success means before an agent spends hours coding. The two files give Codex a stable target, repeatable checks, memory across a long run, and an honest not-ready state when important information is missing.",
    note:
      "In the source workflow, /goal is Codex's long-running task mode. SPEC.md describes the product decision; GOAL.md tells Codex how to execute and verify it; PLAN.md, ATTEMPTS.md, and NOTES.md preserve progress and learning across the run.",
    keywords: [
      "Goal Forge",
      "Codex goal planning",
      "SPEC.md",
      "GOAL.md",
      "autonomous coding contract",
    ],
    related: ["codex-completion-contract-loop", "ticket-to-pr-ready-loop"],
  },
  {
    number: "036",
    slug: "ui-ux-score-loop",
    title: "The UI/UX Score Loop",
    summary:
      "Walks through a real user task, scores each screen, improves weak spots, and retests it.",
    seoTitle: "Browser UI/UX Score Loop | Loop Library",
    description:
      "A browser-based review that completes a real user task, scores each meaningful screen with the same checklist, improves weak spots, and retests the whole task.",
    categoryLabel: "AI frontend design workflow",
    author: "Hayden Cassar (@hcassar93)",
    sourceUrl: "https://github.com/hcassar93/ui-ux-score-loop",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "Improve [user flow, such as signup] at [URL] until [completion criterion]. In a real browser, start each pass from fresh state—no saved login, cookies, or site data. Capture meaningful screens at the agreed sizes and modes, score them with one checklist, and improve the weakest safe area. Rerun the whole flow and keep only regression-free changes. Stop on success, two full passes with no gain, blocked access, or required approval. Return scores, screenshots, changes, and stop reason.",
    verifyTitle:
      "The complete user task scores better without making another important screen worse.",
    verifyDetail:
      "The final dashboard shows the same entry point, fresh browser state, screen sizes, modes, scoring rubric, screenshots, score changes, and stop reason for every retained improvement.",
    useWhen:
      "Use this for a real task such as signup, login, onboarding, checkout, sharing, or creating and editing an item when the entire experience can be exercised in a browser and scored consistently.",
    steps: [
      "Choose the user task, starting URL, success target, browser, clean-session rule, screen sizes, light or dark modes, screens to capture, and anything the agent must not change.",
      "Complete the task once without editing; capture normal screens plus meaningful loading, error, recovery, and success states, then score each with the same user-focused rubric.",
      "Improve the weakest safe area, start a new clean browser session, and repeat the entire task under the same conditions so before-and-after scores are comparable.",
      "Keep only changes that improve the target without hurting another important screen; stop on success, two passes with no gain, blocked access, or required approval.",
    ],
    why:
      "A clean browser session exposes problems that saved logins, cookies, and remembered settings can hide. Repeating the same task with the same scoring rubric makes the result comparable instead of relying on a vague impression that the interface feels better.",
    note:
      "A flow means a user goal, such as signing up or checking out—not a guessed web address. A screen size is sometimes called a viewport; a mode may be light or dark. Judge what the user can see and do, not hidden console output.",
    keywords: [
      "UI UX score",
      "browser flow audit",
      "responsive design review",
      "fresh browser state",
      "user experience improvement",
    ],
    related: ["war-loops-frontend-designer", "full-product-evaluation-loop"],
  },
  {
    number: "037",
    slug: "cold-load-trimmer-loop",
    title: "The cold-load trimmer loop",
    summary:
      "Reduces data downloaded before a web app's first screen without changing behavior or appearance.",
    seoTitle: "Cold-Load Byte Reduction Loop | Loop Library",
    description:
      "A web performance workflow that reduces the data downloaded before the first screen appears, while tests and screenshots protect behavior and appearance.",
    categoryLabel: "AI coding agent workflow",
    author: "Christian Katzmann",
    sourceUrl: "https://github.com/Christian-Katzmann",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "Reduce the data [web app] downloads before its first screen appears. First record passing tests, mobile and desktop screenshots, and compressed transferred bytes—the data actually downloaded. Use the build report only to suggest candidates. Defer, compress, or remove one item, then rebuild and rerun every check. Keep it only if tests pass, screenshots are pixel-identical, and bytes decrease; otherwise revert. Stop when no safe candidate remains, progress stalls, or approval is needed. Return measurements, changes, and untested states.",
    verifyTitle:
      "The first screen downloads less data without a tested behavior or pixel changing.",
    verifyDetail:
      "The same production-like measurement reports fewer downloaded bytes, existing tests pass, every representative screenshot is pixel-identical, and uncertain dependency removal remains approval-gated.",
    useWhen:
      "Use this when a web app feels heavy on its first visit because it downloads too much code, styling, media, or other data before showing the initial screen.",
    steps: [
      "Before changing code, record passing tests, representative mobile and desktop screenshots of the first screen, and a repeatable baseline for compressed transferred bytes—the amount actually downloaded.",
      "Use a build or bundle report to find large or early downloads, then choose one safe candidate to delay until needed, compress, inline, or remove.",
      "Rebuild and rerun the same tests, screenshots, and download measurement; keep the change only when every gate passes and bytes decrease, otherwise revert it completely.",
      "Repeat until no safe candidate remains, several attempts produce no improvement, the measurement is unreliable, or the next change needs approval.",
    ],
    why:
      "Recording behavior and screenshots before the first change prevents a broken screen from becoming the new normal. One download change per round also makes it clear which edit saved bytes and makes a failed attempt easy to undo.",
    note:
      "Measure compressed transferred bytes—the amount sent over the network—not the larger source files developers read. Screenshots protect only the states they capture, so include logged-out, logged-in, empty, error, or other relevant first screens before trusting a risky change.",
    keywords: [
      "first load bytes",
      "bundle size optimization",
      "pixel identical screenshots",
      "lazy loading",
      "web performance loop",
    ],
    related: ["sub-50ms-page-load-loop", "pixel-safe-css-trim-loop"],
  },
  {
    number: "038",
    slug: "pixel-safe-css-trim-loop",
    title: "The pixel-safe CSS trim loop",
    summary:
      "Shrinks styling code sent to users while keeping every tested screen visually identical.",
    seoTitle: "Pixel-Safe CSS Reduction Loop | Loop Library",
    description:
      "A stylesheet cleanup workflow that removes one piece of unused or redundant CSS at a time and keeps it removed only when every tested screen looks identical.",
    categoryLabel: "AI frontend design workflow",
    author: "Christian Katzmann",
    sourceUrl: "https://github.com/Christian-Katzmann",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "Reduce the CSS styling code [site] sends to users without changing tested screens. First capture representative pages, sizes, themes, and interactions, and record the built CSS size. Treat coverage reports only as suggestions. Remove one declaration or rule, rebuild, and rerun screenshots and project checks. Keep it only if every screenshot is pixel-identical and built CSS is smaller; otherwise revert. Stop when no supported candidate remains, progress stalls, or approval is required. Return reduction, evidence, and untested states.",
    verifyTitle:
      "The delivered stylesheet is smaller while every tested screen remains pixel-identical.",
    verifyDetail:
      "The same project checks and screenshots pass after each retained deletion, the built CSS file sent to users is smaller, and untested browsers, screens, or interactions remain explicit risks.",
    useWhen:
      "Use this when a website's styling files may contain unused declarations, duplicated rules, or old overrides and representative pages and interactions can be captured in repeatable screenshots.",
    steps: [
      "Before deleting styling code, list representative pages, screen sizes, light and dark modes, conditional content, hover and keyboard-focus states, and other important variants; capture screenshots and record the final built CSS size.",
      "Use a CSS coverage report to suggest declarations or rules that may be unused, then remove one candidate from the maintainable source file.",
      "Rebuild, run project checks, and recreate every screenshot; keep the deletion only when all images are pixel-identical and the built CSS is smaller, otherwise revert it.",
      "Repeat until no well-supported candidate remains, repeated attempts save nothing, the screenshots cannot cover the affected behavior, or approval is required.",
    ],
    why:
      "Screenshots taken before cleanup preserve the current appearance as the standard. Exact image comparison and one deletion per round catch visual changes that an automated coverage report cannot understand, including rules that matter only because of their order.",
    note:
      "CSS is the code that controls a page's appearance. A coverage report can suggest that a rule was not used during one visit, but it cannot prove the rule is unnecessary everywhere. Add uncertain browsers and interaction states before deleting their styles.",
    keywords: [
      "CSS cleanup",
      "pixel safe CSS",
      "visual regression testing",
      "dead CSS removal",
      "stylesheet optimization",
    ],
    related: ["cold-load-trimmer-loop", "ui-ux-score-loop"],
  },
  {
    number: "039",
    slug: "easy-onboarding-loop",
    title: "The easy onboarding loop",
    summary:
      "Acts like a first-time user, fixes one obstacle, and retries from a completely clean session.",
    seoTitle: "Fresh-State Onboarding Improvement Loop | Loop Library",
    description:
      "A first-time-user test that starts with no saved account or browser state, fixes one confirmed onboarding obstacle, and retries the entire experience.",
    categoryLabel: "AI product evaluation workflow",
    author: "Eric Lott",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "Act like a first-time user of [product]. Start at the real entry point in a clean session with no saved login, site data, remembered route, or hidden setup. Complete onboarding using only visible guidance and record obstacles. Fix the worst one with the smallest change that preserves every security, access, and product requirement. Discard the session and retry. Stop after one uninterrupted success, no safe fix, blocked access, or required approval. Return the path, changes, evidence, and blockers.",
    verifyTitle:
      "A first-time user can complete onboarding in one uninterrupted clean session.",
    verifyDetail:
      "The full experience succeeds from the real starting point without saved browser state, secret setup, guessed routes, or manual repairs, and every real requirement remains intact.",
    useWhen:
      "Use this when new users may face unclear instructions, hidden assumptions, difficult recovery, or unnecessary steps that experienced users no longer notice because their accounts and browsers remember earlier setup.",
    steps: [
      "Open a clean session with no saved login, cookies, site storage, remembered web address, secret setup, or repair left over from an earlier attempt.",
      "Begin where a real newcomer begins, complete the onboarding steps using only visible guidance, and record anything unclear, unexplained, unnecessarily difficult, or impossible to recover from.",
      "Fix the most harmful obstacle with the smallest change that preserves security, access, legal, onboarding, and product requirements.",
      "Throw away the session and retry the entire experience until one uninterrupted clean pass succeeds or no safe progress is possible, access is blocked, or approval is required.",
    ],
    why:
      "Saved logins and remembered setup hide problems from experienced users. Starting over after every fix shows whether the product itself now explains the path, while preserving real requirements prevents an easier experience from weakening security or access controls.",
    note:
      "A clean session means a new private browser or another isolated environment with no cookies, login, local storage, cache, or remembered route. Start where a newcomer would actually arrive and follow only the guidance the product exposes.",
    keywords: [
      "onboarding improvement",
      "fresh session testing",
      "new user experience",
      "agent friendly onboarding",
      "onboarding friction",
    ],
    related: ["fresh-clone-loop", "full-product-evaluation-loop"],
  },
  {
    number: "040",
    slug: "accessibility-repair-loop",
    title: "The accessibility repair loop",
    summary:
      "Finds barriers for keyboard, screen-reader, low-vision, and other users, then fixes the most harmful first.",
    seoTitle: "Accessibility Repair Loop | Loop Library",
    description:
      "An accessibility review that confirms barriers against an agreed standard, fixes the issue with the greatest user impact, and repeats the same checks.",
    categoryLabel: "AI frontend design workflow",
    author: "Eric Lott",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "Check [scope] against [accessibility standard, such as WCAG 2.2 AA] with automated scans and available keyboard, screen-reader, and other manual tests. Confirm each issue, rank it by harm, and fix the highest-impact blocker. Rerun the same checks, affected task, and regression tests. Keep only verified fixes. Stop when no blocker remains, progress stalls, verification is unavailable, or approval is required. Never silence a check or weaken the target. Return issues, fixes, evidence, exceptions, and untested needs.",
    verifyTitle:
      "No confirmed accessibility barrier remains in the agreed pages, components, or user tasks.",
    verifyDetail:
      "The same automated scans, available manual checks, affected user task, and regression tests pass after each retained fix without lowering the chosen accessibility standard.",
    useWhen:
      "Use this when a website or app has a defined accessibility target and you can repeatedly test the relevant pages, components, or tasks for people using keyboards, screen readers, zoom, or other access methods.",
    steps: [
      "Choose the pages, components, and user tasks to test; name the accessibility standard, such as WCAG 2.2 AA; and list the automated scans and manual checks that are actually available.",
      "Run the baseline, reproduce each finding instead of trusting a tool warning by itself, and rank confirmed barriers by the number of people affected and how severely they are blocked.",
      "Fix the most harmful barrier with the smallest underlying change, then repeat the same scan, manual check, user task, and relevant regression tests.",
      "Keep only verified fixes and repeat until no confirmed barrier remains or progress stalls, evidence cannot be collected, work is blocked, or approval is required.",
    ],
    why:
      "A fixed scope and repeated checks keep accessibility work tied to real people and reproducible evidence instead of an endless score chase. Fixing the most harmful confirmed barrier first directs effort to the users who are blocked most severely.",
    note:
      "Automated tools can find likely problems but cannot prove a product is accessible. Manual keyboard use, screen-reader checks, zoom, contrast review, and real user testing may still be needed. Record anything the available test setup could not cover.",
    keywords: [
      "accessibility audit",
      "accessibility repair",
      "WCAG workflow",
      "inclusive design testing",
      "accessibility regression",
    ],
    related: ["ui-ux-score-loop", "full-product-evaluation-loop"],
  },
  {
    number: "041",
    slug: "housekeeper-loop",
    title: "The housekeeper loop",
    summary:
      "Cleans a code project one proven, low-risk change at a time without touching uncertain work.",
    seoTitle: "Repository Housekeeper Cleanup Loop | Loop Library",
    description:
      "A conservative code-project cleanup that proves one small opportunity is safe, makes the smallest useful change, and keeps it only after existing checks pass.",
    categoryLabel: "AI coding agent workflow",
    author: "Eric Lott",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "Review [repository or code project] for dead code, meaning unreachable or unused code; stale files or comments; unused dependencies; duplication; broken links; inconsistent names; and confusing structure. Protect unrelated, active, uncommitted, generated, and uncertain work. Prove one low-risk cleanup, make the smallest coherent change, then rerun the build, tests, runtime checks, and diff review. Keep only verified improvements. Stop when none remain, progress stalls, verification is unavailable, or approval is required. Return changes, evidence, and deferred candidates.",
    verifyTitle:
      "No confirmed low-risk cleanup remains, and existing behavior still passes.",
    verifyDetail:
      "Every retained cleanup is supported by direct evidence, relevant builds and tests pass, the application still runs where applicable, unrelated work is untouched, and uncertain candidates are deferred rather than deleted.",
    useWhen:
      "Use this when a code project has accumulated small maintenance problems—unused code, stale files, duplicated logic, broken links, old comments, inconsistent names, or confusing organization—but broad deletion would be risky.",
    steps: [
      "Inspect the current code-project state and identify active branches, uncommitted edits, generated files, configuration, and other work that must not be disturbed.",
      "Collect possible cleanups, then use code references, configuration, documentation, tests, or runtime behavior to prove one candidate is genuinely low risk.",
      "Make the smallest useful change and run the existing build, tests, application checks, and diff review; keep it only when behavior stays intact and no unrelated work changes.",
      "Repeat until no confirmed low-risk cleanup remains, progress stalls, verification is unavailable, or the next candidate needs approval.",
    ],
    why:
      "One proven cleanup at a time keeps the work easy to review and undo. Requiring evidence before deletion—and protecting uncertain files and edits—prevents a tidy-up pass from removing code that is active but poorly understood.",
    note:
      "Here, repository means the code project and its files. This loop cleans source and project structure; the separate repository cleanup loop handles Git branches, pull requests, commits, and worktrees. Never delete something merely because its purpose is not immediately obvious.",
    keywords: [
      "codebase housekeeping",
      "dead code cleanup",
      "unused dependency review",
      "repository hygiene",
      "incremental cleanup",
    ],
    related: ["repository-cleanup-loop", "overnight-docs-sweep"],
  },
  {
    number: "042",
    slug: "axelrod-subagent-arena-loop",
    title: "The Axelrod subagent arena loop",
    summary:
      "Tests whether AI agents learn to cooperate, retaliate, or forgive in a repeated two-choice game.",
    seoTitle: "Axelrod Subagent Arena Benchmark | Loop Library",
    description:
      "A controlled tournament where two reasoning AI agents repeatedly choose to cooperate or defect, then are compared with players that always make one choice.",
    categoryLabel: "AI product evaluation workflow",
    author: "Kan Yuenyong (@sikkha)",
    sourceUrl: "https://github.com/sikkha/axelrod-loop-engineering",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "Run a fixed Axelrod tournament with two reasoning AI agents. Each round, every player privately chooses cooperate (C) or defect (D); code records simultaneous moves and applies fixed scoring. Include always-defect and always-cooperate comparison players. Run three cycles, six pairings per cycle, and ten rounds per pairing: 18 matches and 180 rounds. Hide opponent type and private reasoning. Validate every move and total. Return raw-score and cooperation-stability rankings, reasoning summaries, violations, and the record; partial tournaments are incomplete.",
    verifyTitle:
      "All 18 matches and 180 rounds can be reproduced from the recorded moves and fixed scoring rules.",
    verifyDetail:
      "Each agent chooses before seeing the opponent's move, every move is recorded before scoring, totals reproduce from the full history, invalid responses are logged, and any partial or invalid tournament remains explicitly incomplete.",
    useWhen:
      "Use this as a controlled experiment to see whether AI agents learn repeated-interaction behaviors such as cooperation, retaliation after betrayal, forgiveness, exploitation, and different strategies for different opponents.",
    steps: [
      "Set up fixed scoring, move validation, the match schedule, stored history for each pair, two reasoning AI players, one player that always cooperates, and one that always defects; code may score moves but never choose for the reasoning agents.",
      "Before each of three tournament cycles, have each reasoning agent choose a bounded strategy using only what happened in its own earlier matches with each opponent.",
      "Run all six possible pairings for ten rounds, collecting cooperate or defect choices simultaneously while hiding opponent identity and private reasoning; record every move, score, and allowed explanation.",
      "Recalculate all 18 matches and 180 rounds from the saved record, then report both total points and cooperation-stability measures, strategy changes, reasoning summaries, rule violations, and any incomplete data.",
    ],
    why:
      "The always-cooperate and always-defect players provide simple comparison points: they reveal whether the reasoning agents exploit easy opponents, defend themselves, rebuild cooperation, or change strategy. Hidden identities, simultaneous choices, saved pair histories, and recalculated scores keep the experiment fair and auditable.",
    note:
      "The scoring rule is: both cooperate, 3 points each; one defects, the defector gets 5 and the cooperator gets 0; both defect, 1 point each. Total-points ranking rewards points earned, while cooperation-stability measures reward reciprocal cooperation, effective retaliation, forgiveness, and resistance to exploitation.",
    keywords: [
      "Axelrod tournament",
      "Iterated Prisoner's Dilemma",
      "multi agent benchmark",
      "agent cooperation",
      "reasoning subagent evaluation",
    ],
    related: ["boeing-747-benchmark", "full-product-evaluation-loop"],
  },
  {
    number: "043",
    slug: "prepare-new-project-loop",
    title: "The prepare-a-new-project loop",
    summary:
      "Strengthens project documents until independent engineers would build substantially the same system.",
    seoTitle: "Prepare a New Project Documentation Loop | Loop Library",
    description:
      "A planning workflow that closes documentation gaps until requirements, technical design, acceptance criteria, and test strategy describe one buildable system.",
    categoryLabel: "AI coding agent workflow",
    author: "Brad Shannon (@bradshannon)",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "Prepare [project] for implementation. Ensure its documents cover requirements, technical design, tasks with acceptance criteria, and test strategy. Each round, fix the largest gap or contradiction that could make two competent engineers build different systems. Keep details traceable, record assumptions, and ask before product forks. Recheck consistency, then have two independent reviewers describe the components, data model, dependencies, and definition of done. Stop when they materially agree and every artifact is testable, or a decision needs the user.",
    verifyTitle:
      "Two independent reviewers derive substantially the same build from the project documents.",
    verifyDetail:
      "Their descriptions agree on the components, data model, dependencies, and definition of done, and every required artifact is specific, consistent, traceable, and testable.",
    useWhen:
      "Use this before building a new software project when its idea or early documents still leave important implementation decisions open to interpretation.",
    steps: [
      "Inventory the current project documents and identify the missing requirements, technical design, task breakdown, acceptance criteria, or test strategy needed before implementation.",
      "Find the single largest gap, contradiction, or vague requirement that could make competent engineers build different systems, then close it with concrete detail traceable to a stated requirement.",
      "Record assumptions that can be made safely, ask the user about genuine product forks, and recheck every edited document against the others for consistency.",
      "Have two independent reviewers describe the intended components, data model, dependencies, and definition of done; repeat until their descriptions materially agree or a required decision blocks progress.",
    ],
    why:
      "A concrete convergence test exposes ambiguity that a single author may read past. Fixing one divergence at a time keeps the documents coherent and turns project preparation into evidence that another engineer can follow rather than a pile of planning text.",
    note:
      "Do not add detail merely to make the documents longer or invent product requirements to force agreement. Keep every claim tied to a stated requirement, record assumptions, and return unresolved product choices to the user.",
    keywords: [
      "project planning loop",
      "build ready documentation",
      "technical design review",
      "requirements convergence",
      "software project preparation",
    ],
    related: [
      "goal-forge-loop",
      "multi-llm-convergence-loop",
      "codex-completion-contract-loop",
    ],
  },
  {
    number: "044",
    slug: "test-stabilizer-loop",
    title: "The test stabilizer loop",
    summary:
      "Finds flaky tests, fixes their root causes, and proves stability with repeated full-suite runs.",
    seoTitle: "Flaky Test Stabilizer Loop | Loop Library",
    description:
      "A flaky-test repair workflow that measures inconsistent results, fixes one root cause at a time, and stops after a defined streak of stable full-suite runs.",
    categoryLabel: "AI coding agent workflow",
    author: "hungtv27 (@hungtv27)",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "Run [test suite] [N] times under the same conditions and list tests whose result changes. Fix the most frequent flake at its root cause—shared state, timing, ordering, or an external dependency—never with a blind sleep or retry. Run that test [N] times, then rerun the full suite. Repeat until [N] consecutive full-suite runs pass, progress stalls, or approval is required. Return each flake, root cause, fix, evidence, and justified quarantine.",
    verifyTitle:
      "The full test suite passes for the required consecutive-run streak.",
    verifyDetail:
      "The repaired test passes repeatedly, [N] consecutive full-suite runs are green under the recorded conditions, and no blind sleep or retry hides an unresolved cause.",
    useWhen:
      "Use this when a test suite produces inconsistent results across otherwise comparable runs and the failures may come from shared state, timing, ordering, or external dependencies.",
    steps: [
      "Choose the test suite, the required run count, and the conditions that must stay fixed, then run the complete suite repeatedly and record every inconsistent test.",
      "Select the most frequent flake, reproduce it as narrowly as practical, and identify the underlying shared-state, timing, ordering, or dependency failure.",
      "Fix the test or product code without adding a blind sleep or retry, then run the affected test repeatedly before returning to the complete suite.",
      "Repeat until the required number of consecutive full-suite runs pass, progress stalls, or approval is needed, and report every root cause, fix, quarantine, and remaining blocker.",
    ],
    why:
      "Repeated runs turn intermittent failures into measurable evidence. Repairing the most frequent flake first and requiring a full-suite streak prevents a local fix from hiding another source of instability.",
    note:
      "Choose [N] before the first run and keep the environment comparable. Quarantine is a visible temporary containment step, not proof of repair; record its reason and do not report the suite as fully stabilized while unresolved tests remain quarantined.",
    keywords: [
      "flaky test repair",
      "test suite stabilization",
      "intermittent test failures",
      "test reliability loop",
      "root cause testing",
    ],
    related: ["quality-streak-loop", "test-suite-speed-loop"],
  },
  {
    number: "045",
    slug: "artifact-to-skill-loop",
    title: "The artifact-to-skill loop",
    summary:
      "Extracts the method behind a strong artifact and proves it works on a fresh case.",
    seoTitle: "Artifact-to-Skill Extraction Loop | Loop Library",
    description:
      "A reusable workflow for turning one proven artifact into a transferable skill, playbook, or procedure and validating it on a second case.",
    categoryLabel: "AI workflow design workflow",
    author: "Hiten Shah (@hnshah)",
    published: "2026-06-20",
    modified: "2026-06-20",
    prompt:
      "Turn [artifact] into a skill, playbook, or procedure. Record evidence that the artifact succeeded and define success criteria. Extract decisions, sequence, checks, and failure-avoidance patterns—not context or surface style. Remove sensitive material. Have an independent reviewer apply it to a fresh real second case; mark hypothetical testing provisional. Revise at most twice. Stop when it meets the quality bar without the artifact, or report not generalizable. Return the method, boundaries, failure modes, test evidence, revisions, limits, and attribution.",
    verifyTitle:
      "The extracted method succeeds on a fresh second case without the original artifact.",
    verifyDetail:
      "An independent reviewer applies the reusable version under criteria defined before extraction, and the second result meets the source artifact's demonstrated quality bar or the method is honestly marked provisional or not generalizable.",
    useWhen:
      "Use this when a completed artifact has evidence of success, appears to contain a repeatable method, and similar work is likely to recur.",
    steps: [
      "Confirm that the source artifact has credible evidence of success, define the quality criteria it met, and exclude sensitive or proprietary material that should not be transferred.",
      "Separate the durable decisions, sequence, checks, standards, and failure-avoidance patterns from one-off facts, tools, and surface style.",
      "Write the method as a standalone skill, playbook, or procedure with inputs, boundaries, steps, quality standards, failure modes, attribution, and clear terminal states.",
      "Have an independent reviewer apply it to a fresh real case, revise no more than twice, and return either a reusable version with test evidence or an honest provisional, blocked, or not-generalizable result.",
    ],
    why:
      "Strong outputs often get saved while the method that produced them disappears. Extracting the decisions and checks makes that knowledge reusable, while a fresh second-case test distinguishes a transferable process from imitation of one polished example.",
    note:
      "Do not infer success from polish alone, copy confidential material, or treat a hypothetical test as final proof. Preserve attribution, define the quality bar before extraction, and stop honestly when hidden context makes the method impossible to generalize.",
    contributorPlaybook: {
      whenNotToUse: [
        "The artifact is mediocre, unclear, or lacks credible evidence that it succeeded.",
        "Essential context is hidden or unavailable, so the method cannot be recovered responsibly.",
        "The artifact is too narrow to generalize beyond its original case.",
        "A summary is sufficient and no reusable method is needed.",
        "The goal is to imitate surface style rather than understand the process behind it.",
      ],
      expectedOutputs: [
        "Artifact summary",
        "Extracted method",
        "Reusable skill, playbook, or procedure",
        "Usage boundaries",
        "Step-by-step procedure",
        "Quality standards",
        "Failure modes",
        "Second-case test",
        "Revisions made after testing",
        "Final reusable version",
      ],
      implementationGuidance: [
        "Prompt or workflow: preserve the sequence, checks, and escalation logic.",
        "Page or memo: preserve the argument structure, evidence logic, and clarity standards.",
        "Pull request or specification: preserve scope control, proof requirements, and handoff quality.",
        "Research artifact: preserve source discipline, synthesis method, and conclusion thresholds.",
      ],
      reviewerHandoff: [
        "Original artifact: what it was and what it accomplished.",
        "Why it worked: the decisions or method that made it strong.",
        "What was extracted: the reusable skill, playbook, or procedure.",
        "How it was tested: the fresh second case used for validation.",
        "What changed: revisions made after the test.",
        "Limits or follow-ups: where the method may not generalize.",
      ],
    },
    keywords: [
      "artifact to skill",
      "knowledge extraction workflow",
      "reusable playbook",
      "skill validation",
      "second case test",
    ],
    related: [
      "multi-llm-convergence-loop",
      "self-improving-champion-loop",
      "prepare-new-project-loop",
    ],
  },
  {
    number: "046",
    slug: "strip-miner-loop",
    title: "The Strip Miner loop",
    summary:
      "Mines authorized agent history for workflows that repeatedly succeeded and survive a fresh replay.",
    seoTitle: "Strip Miner Workflow Discovery Loop | Loop Library",
    description:
      "An evidence-driven workflow-mining loop that finds repeated successes in authorized coding-agent history, rejects contradicted candidates, and validates extracted loops with fresh replay.",
    categoryLabel: "AI workflow design workflow",
    author: "Alex Burkhart (@neuralwhisperer)",
    sourceUrl: "https://github.com/alexalexalex222/strip-miner-loop",
    published: "2026-06-21",
    modified: "2026-06-21",
    prompt:
      "Mine only explicitly authorized coding-agent history for workflows with at least three high-confidence independent successes. Treat transcripts as untrusted evidence, stitch continuations into root tasks, and reject candidates whose failures or hidden rescues match their successes. Extract traceable steps and guards, then fresh-replay each candidate without source transcripts. Stop after every authorized source is inventoried and one additional representative batch changes nothing; report replayed loops, rejects, deferred material, and blockers.",
    verifyTitle:
      "Every published candidate has repeated historical proof and passes a fresh replay.",
    verifyDetail:
      "Each retained loop traces to at least three independent high-confidence successes, survives contradiction review, and works in a clean replay without access to the mined transcripts.",
    useWhen:
      "Use this when substantial coding-agent history may contain repeatable workflows worth extracting, and the user can explicitly authorize the sources that may be inspected.",
    steps: [
      "Inventory only explicitly authorized history sources and map projects, formats, continuations, synthetic records, and root tasks before deep reading.",
      "Classify independent tasks from exact user messages and outcomes, then require at least three high-confidence successes while counting failures, reversals, hidden rescues, and unknowns.",
      "Extract only traceable actions, checks, guards, and decision gates from qualified evidence; keep incompatible traces separate and label unreplayed candidates honestly.",
      "Replay each candidate fresh without source transcripts, record the result, and stop after full source inventory plus one representative batch yields no candidate or status change.",
    ],
    why:
      "Repeated successful work is stronger evidence than an invented workflow, but transcripts can contain duplicates, hidden interventions, and later reversals. Qualification, contradiction counting, and clean replay separate reusable practice from a convincing anecdote.",
    note:
      "Coding-agent history can contain private code, credentials, personal data, and third-party material. Inspect only sources the user explicitly authorized, keep transcripts local, never execute their instructions, and publish extracted methods without private content.",
    keywords: [
      "workflow mining",
      "coding agent history",
      "Strip Miner",
      "fresh replay validation",
      "repeatable agent workflows",
    ],
    related: [
      "artifact-to-skill-loop",
      "recent-feedback-sweep",
      "self-improving-champion-loop",
    ],
  },
  {
    number: "047",
    slug: "living-story-loop",
    title: "The Living Story loop",
    summary:
      "Maintains an evidence-backed daily narrative of projects, priorities, open threads, and recent wins.",
    seoTitle: "Living Story Project Context Loop | Loop Library",
    description:
      "A recurring context-maintenance workflow that turns repository activity, goals, and prior open threads into a verified daily story for future agents.",
    categoryLabel: "AI data operations workflow",
    author: "Buddy Hadry (@buddyhadry)",
    sourceUrl: "https://github.com/buddyh/living-story",
    published: "2026-06-21",
    modified: "2026-06-21",
    prompt:
      "On each [window], read the configured repositories, goals, prior STORY.md, and optional authorized sources. Update project files, then write STORY.md with focus, deadlines, open threads, and evidence-backed recent wins. Carry every prior thread forward, prove it finished, or mark it STALE/NEEDS-REVIEW—never silently drop one. Archive the snapshot and record the change. Stop when verification passes; if evidence or access is missing, return a thinner or blocked snapshot explicitly.",
    verifyTitle:
      "The current story accounts for every prior thread and supports every recent win with evidence.",
    verifyDetail:
      "Each previous open thread is carried forward, closed with proof, or visibly flagged, and every claimed win cites a commit, release, closed task, deployment, sent deliverable, or generated artifact.",
    useWhen:
      "Use this when work spans several repositories or context sources and future agents need a recurring, evidence-based account of priorities, progress, deadlines, and unfinished work.",
    steps: [
      "Read the configured repositories, goals, personal context, optional authorized sources, previous STORY.md, and existing project files; report missing inputs instead of inventing them.",
      "Refresh each project record with current activity, branch state, shipped evidence, in-progress work, and stale status under the configured window.",
      "Write the new story with interpretation, focus, deadlines, open threads, and evidence-backed recent wins rather than a raw commit list.",
      "Reconcile every previous thread, archive the verified snapshot, update the changelog, and stop with an explicit complete, thinner, or blocked result.",
    ],
    why:
      "A recurring narrative preserves the meaning behind activity without letting old commitments disappear. Evidence requirements keep recent wins factual, while thread reconciliation makes stale or unfinished work visible to the next agent.",
    note:
      "Configure source paths and the stale window before relying on the story. Treat notes, calendars, task exports, and repository history as private; read only authorized sources and do not publish or transmit their contents without approval.",
    keywords: [
      "Living Story",
      "agent context management",
      "project status narrative",
      "open thread tracking",
      "evidence based progress",
    ],
    related: [
      "five-minute-repository-maintainer-loop",
      "nightly-changelog-sweep",
      "recent-feedback-sweep",
    ],
  },
  {
    number: "048",
    slug: "groundtruth-audit-loop",
    title: "The Groundtruth loop",
    summary:
      "Audits a project from direct evidence and reports every area as proved, weak, or unverified.",
    seoTitle: "Groundtruth Evidence-Based Project Audit | Loop Library",
    description:
      "A read-only project-audit workflow that verifies architecture, security, platform behavior, operations, and business logic from current evidence rather than assumptions.",
    categoryLabel: "AI coding agent workflow",
    author: "Mohamed (@aivibecode)",
    published: "2026-06-21",
    modified: "2026-06-21",
    prompt:
      "Audit [project] from its actual code and configuration, not framework assumptions. For architecture, platform compatibility, security, privileged areas, performance, deployment, jobs, business logic, and code quality, record proved, no issue, weak, or N/A with direct evidence; verify external limits from current primary sources and calculate numbers. Ask before changing code. Stop when every area is logged with severity, or return unverified areas as blocked. Finish with a plain-language overview and area-to-evidence table.",
    verifyTitle:
      "Every audit area has a current evidence-backed outcome and severity.",
    verifyDetail:
      "The area-to-evidence table contains no silent gaps: each area is proved, no issue found, weak, N/A with a reason, or explicitly unverified and blocked.",
    useWhen:
      "Use this before trusting a project's security, correctness, platform compatibility, privileged surfaces, scheduled work, or operational assumptions and when the first task is audit rather than repair.",
    steps: [
      "Discover the real language, framework, hosting platform, privileged surfaces, scheduled jobs, and deployment configuration from the scoped project itself.",
      "Inspect each required area, tie conclusions to code or configuration, verify platform and library behavior from current primary sources, and calculate rather than estimate quantitative claims.",
      "Record an outcome, evidence, and severity for every area, separating confirmed weaknesses from no-issue findings, justified N/A results, and unverified gaps.",
      "Deliver the plain-language project overview and area-to-evidence table without changing code; stop complete only when every area is accounted for, otherwise return the blocked gaps.",
    ],
    why:
      "Broad audits fail when they inherit framework defaults, rely on remembered limits, or omit quiet areas. A fixed evidence table forces the reviewer to prove, clear, exclude, or explicitly block every surface.",
    note:
      "This loop is read-only. Ask before changing code, configuration, infrastructure, or production state. Use current primary documentation for external behavior, avoid exposing secrets from privileged areas, and do not turn missing access into a clean finding.",
    keywords: [
      "Groundtruth audit",
      "evidence based code review",
      "project security audit",
      "platform compatibility review",
      "area to evidence table",
    ],
    related: [
      "full-product-evaluation-loop",
      "promise-to-proof-loop",
      "recent-feedback-sweep",
    ],
  },
  {
    number: "049",
    slug: "recovery-proof-loop",
    title: "The Recovery Proof loop",
    summary:
      "Proves real backups can restore required scenarios inside a disposable clean-room environment.",
    seoTitle: "Backup Recovery Proof Loop | Loop Library",
    description:
      "A disaster-recovery validation workflow that restores randomly selected real recovery points, verifies integrity and RPO/RTO, and preserves failures as regression drills.",
    categoryLabel: "AI recovery operations workflow",
    author: "Eric Lott",
    published: "2026-06-21",
    modified: "2026-06-21",
    prompt:
      "For each required recovery scenario, randomly select an eligible real backup or recovery point and restore from zero in a disposable, isolated clean-room using only documented materials. Verify integrity, dependencies, representative reads and writes, and actual RPO and RTO. Repair one blocker, destroy the environment, and retry fresh. Stop when every scenario reaches its predefined consecutive-success streak or an exception is explicitly accepted. Never overwrite production, expose restored data, or initiate failover without approval.",
    verifyTitle:
      "Every required recovery scenario succeeds repeatedly from a real recovery point.",
    verifyDetail:
      "Fresh clean-room restores satisfy integrity, dependency, representative read/write, RPO, and RTO checks under unchanged criteria, with failures preserved as regression drills and restored data destroyed securely.",
    useWhen:
      "Use this when backup existence is not enough and the organization needs repeatable proof that required systems can be restored from documented materials within agreed recovery objectives.",
    steps: [
      "Define the required scenarios, eligible recovery points, unchanged success criteria, consecutive-success streak, isolation controls, and approval boundaries before restoring anything.",
      "Randomly select one eligible real recovery point, restore from zero in a disposable clean-room using only documented materials, and measure actual RPO and RTO.",
      "Verify checksums, control totals, referential integrity, keys, dependencies, and representative business reads and writes; preserve any failure as a regression drill.",
      "Repair one recovery blocker, destroy the environment securely, and retry fresh until every scenario passes its streak or an unresolved exception is explicitly accepted.",
    ],
    why:
      "A backup is only useful if a real recovery point can rebuild the required system under documented conditions. Random selection, fresh environments, measured objectives, and repeated success expose gaps that a one-time scripted restore can hide.",
    note:
      "Restored production data remains sensitive even in a test environment. Never overwrite production, weaken isolation, expose restored data, or initiate production failover without explicit approval; preserve immutable evidence and securely destroy test data after each run.",
    keywords: [
      "backup recovery testing",
      "disaster recovery drill",
      "RPO and RTO validation",
      "clean room restore",
      "recovery proof",
    ],
    related: [
      "quality-streak-loop",
      "post-release-baseline-loop",
      "production-error-sweep",
    ],
  },
  {
    number: "050",
    slug: "refund-follow-up-loop",
    title: "The refund follow-up loop",
    summary:
      "Keeps pursuing a refund until the money arrives or the agent genuinely needs the user.",
    seoTitle: "Refund Follow-Up Loop for AI Agents | Loop Library",
    description:
      "A persistent follow-up workflow that starts a refund claim, watches replies and deadlines, and keeps the case moving until the money arrives.",
    categoryLabel: "AI consumer advocacy workflow",
    author: "Jason (@jxnlco)",
    sourceUrl: "https://x.com/jxnlco",
    published: "2026-06-21",
    modified: "2026-06-21",
    prompt:
      "Get my refund for [company and charge info]. Start the claim now through an approved support channel, then keep following up on replies, promises, and deadlines until the refund arrives. Keep a short case note so each follow-up has context. Stop only when the refund is received or you are genuinely blocked and need me.",
    verifyTitle:
      "The refund is received, or a genuine blocker requires the user.",
    verifyDetail:
      "An open claim, promise, or pending refund is progress, not success; keep following up until the money arrives or no approved next step remains.",
    useWhen:
      "Use this when someone owes you a refund and getting it may take more than one support conversation or follow-up.",
    steps: [
      "Gather the charge, reason for the refund, useful evidence, current status, and any earlier conversation or promise.",
      "Start or continue the claim through a support channel the user has approved, then note what happened and what should happen next.",
      "Follow up whenever a reply, promise, or deadline creates a useful next step; keep the case moving instead of treating a pending status as done.",
      "Stop when the refund arrives, or explain the genuine blocker when the next useful step needs the user.",
    ],
    why:
      "Refunds often stall because a promise or pending status gets treated as completion. This loop keeps ownership through delays and handoffs until the money actually arrives.",
    note:
      "Use truthful information and the permissions already granted. If the next step needs a new permission or decision, bring that blocker to the user instead of stopping silently.",
    keywords: [
      "refund follow up",
      "consumer advocacy",
      "customer support escalation",
      "refund status tracking",
      "case log",
    ],
    related: [
      "promise-to-proof-loop",
      "living-story-loop",
      "recent-feedback-sweep",
    ],
  },
];

validateLoopData(loops);