All notable changes to this plugin will be documented here.
Format: Added / Changed / Fixed / Removed per release.
- Redesigned the plugin from a token-heavy 10-skill / 6-agent model to a lightweight 3-skill / 0-agent model focused on high-quality plan generation, inline adversarial plan verification, and evidence-backed review layered over the native Claude Code toolset.
controlflow-planis now schema-sourced: it reads the sharedschemas/planner.plan.schema.jsonandplans/templates/plan-document-template.mdat invoke time instead of carrying a frozen planning body.controlflow-reviewis now a thin layer over native/code-review/security-review: delegates the mechanical/style pass to native review and adds plan-vs-implementation scope drift, evidence discipline, and proactive vulnerability/error search.- Manifest (
plugin.json) and marketplace entry bumped to 0.2.0 with an updated description and keywords reflecting the standalone, zero-subagent design. - Plugin decoupled from the shared-source generator:
claude_coderemoved fromgeneration-manifest.jsontargets and fromsync-plugin-assets.ps1/validate-generated-assets.ps1host sets. This plugin is now a standalone, hand-maintained source-of-truth for the Claude Code host; codex and cursor remain generator-managed. - Repo
CLAUDE.mdslimmed to a routing stub (~64% smaller) that preserves all drift anchors (YAML header withAgent: Planner+Schema Version: 1.2.0, the "below 0.9" confidence threshold phrase, the 8-agent executor enum sentence, and the 7 semantic-risk categories).
controlflow-verifyskill: three-phase inline adversarial verification (structural audit, assumption/mirage check, executability cold-start) with tier gating and an APPROVED / NEEDS_REVISION / REJECTED verdict; writesplans/artifacts/<task-slug>/verify-verdict.md. Adversarial framing and anti-rationalization rules built into the skill so inline verification is not a rubber stamp.- Reference docs for the new skills:
plan-format.md,complexity-tiers.md,inline-execution.md,llm-behavior-guidelines.md(plan);adversarial-framing.md,verify-phases.md,mirage-patterns.md(verify);review-checklist.md,validation-status.md,evidence-discipline.md,security-review-discipline.md(review).
- Nine obsolete skills:
controlflow-router,controlflow-spec,controlflow-strict-workflow,controlflow-planning,controlflow-plan-audit,controlflow-assumption-verifier,controlflow-executability-verifier,controlflow-orchestration,controlflow-memory-hygiene. - Six obsolete plugin agents:
controlflow-code-mapper,controlflow-researcher,controlflow-plan-auditor,controlflow-assumption-verifier-agent,controlflow-executability-verifier-agent,controlflow-code-reviewer. Theagents/directory is gone; verification runs inline. - Orphaned
host-overrides/claude-code/override directory (no longer generator-managed). - V1 plugin-local tooling that encoded the deleted verifier/audit model: the three report templates (
assumption-verifier-report-template.md,executability-verifier-report-template.md,plan-audit-report-template.md), thescripts/validate-claude-artifacts.ps1validator, and thetests/directory (fixtures +validate-claude-artifacts.test.ps1).controlflow-verifywrites a single combinedverify-verdict.md, so the per-verifier report templates no longer apply. Dependent references updated in lockstep:plugins/controlflow-shared-source/tests/validate-generated-assets.test.ps1(TEST 4 removed),plugins/controlflow-shared-source/README.md(claude-code dropped from generated-output / host-override lists),CONTRIBUTING.md(claude-code section now points to the eval suite +USAGE.md), anddocs/architecture-decisions/0002-planner-schema-2.0.0-roadmap.md(claude-code fixtures dropped from the 2.0.0 migration inventory).
- None of the removed tooling was referenced by the repo
evals/suite or CI (ci.ymlruns onlycd evals && npm test, which executes the.mjsgate); removal does not affect the eval gate.
- Plugin skeleton under plugins/controlflow-claude-code/
- Manifest at .claude-plugin/plugin.json with name, version, description, author, repository, license, and keywords
- README.md and USAGE.md initial shells
- Report templates: plan-audit-report-template.md, assumption-verifier-report-template.md, executability-verifier-report-template.md
- Planning reference files: plan-template.md, complexity-tiers.md, semantic-risk-taxonomy.md, controlflow-portability.md, planner-output-contract.md
- Skills (Phase 3): ten ControlFlow workflow skills adapted for Claude Code native slash invocation
- Agents (Phase 4): six selected plugin agents for isolated audit, research, and review work (including
controlflow-assumption-verifier-agentandcontrolflow-executability-verifier-agent) - Validator and tests (Phase 5): validate-claude-artifacts.ps1, test suite, and fixtures
- Documentation (Phase 6): README.md, USAGE.md, and integrated project docs