chore(memory): seeder for global dev-prohibition rules into Ralph recall#26
Merged
Merged
Conversation
… into Ralph recall
Adds scripts/memory/seed-dev-prohibitions.sh — a one-command wrapper that
graduates the 4 global development-prohibition rules into the project-scoped
Ralph Memory Tree (recall_v2 / Codex recall):
1. no placeholders in any development
2. no fallbacks unless explicitly requested (fail loud / fail fast)
3. no production code added solely to make a test pass
4. unit AND e2e tests must fail loud / fail fast (special force for heavy
e2e on local minikube: never skip/mock/pretend success)
The wrapper runs migrate_rules_to_nodes.py (GREEN nodes, confidence 1.0) then
project_memory.py to refresh the read-only MEMORY.md projection, and finally
verifies via recall_v2 that all 4 nodes are surfaced — failing loud if any is
missing (it follows the very rules it seeds: fail-fast, no silent fallbacks).
Idempotent: deterministic node ids update in place, never duplicate. The Memory
Tree is project-scoped by design (recall_v2 hard-rejects foreign project_ids),
so run this per-repo where in-repo Codex recall is wanted. Global coverage for
all projects is provided separately by ~/.claude/CLAUDE.md and
~/.claude/rules/proven/*.md (outside this repo).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 15cbb7a6e5
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| fi | ||
|
|
||
| echo "==> Migrating dev-prohibition rules into the Ralph Memory Tree ($MODE)" | ||
| python3 "$MIGRATE" --rules "$SEED_FILE" "$MODE" |
There was a problem hiding this comment.
On a first run where ~/.ralph/procedural/seed-dev-prohibitions.json exists but ~/.ralph/cache/ has not been created yet, this apply path fails after the nodes are processed: migrate_rules_to_nodes.py writes rejects_path.parent / "migrated-nodes.jsonl" without creating the parent when there are no rejects. The four expected rules are valid, so this common path raises FileNotFoundError instead of completing the one-command seeding; ensure the cache directory exists or pass a rejects path whose parent is created before calling the migrator.
Useful? React with 👍 / 👎.
|
|
||
| if [[ "$MODE" == "--apply" ]]; then | ||
| echo "==> Refreshing the read-only GREEN projection in MEMORY.md" | ||
| python3 "$PROJECT_MEMORY" --apply |
There was a problem hiding this comment.
If the user has configured the dedicated memory-tree location with RALPH_MEMORY_HOME, the migration above writes via TreeStore() into that directory, but this projection command (and the recall verification below) default to RALPH_HOME/~/.ralph instead. In that environment the script successfully creates the nodes in one tree, then projects zero nodes and fails verification from another tree; pass the resolved Ralph memory home consistently to migrate_rules_to_nodes.py, project_memory.py, and recall_v2.py.
Useful? React with 👍 / 👎.
…ture The 'Validate hooks structure' gate required every file under .claude/hooks/ to end in .sh, but the repo legitimately ships polyglot hooks (git-safety-guard.py, audit-secrets.js, context-mode-cache-heal.mjs). This made the check fail on main and on every PR. Accept .sh/.py/.js/.mjs and still fail loudly on any unsupported extension. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replace the legacy ZeroLeaks multi-agent security description (419 lines) with a narrowed adversarial opposite-analysis skill (102 lines): claim challenge workflow, assumption testing, counterargument generation, failure-mode discovery, external advisor routing (Z.ai/MiniMax via the Ralph MCP router), security mode, output shapes, and evidence-grounded guardrails. Codex main stays final owner of decisions and edits. Keeps VERSION frontmatter (bumped 3.0.0 -> 3.1.0) required by the v2.87 skills unification validator. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Address review findings on the rewritten adversarial skill: - Make it harness-agnostic: replace "Codex main" owner framing with "main agent / orchestrator" so a Claude Code-loaded skill no longer names a different agent as decision owner. - Mechanical sensitivity gate: external routing now MUST scan the full brief with scripts/memory/sensitive_content.py (classify_text); GREEN only, scanner-authoritative, never a self-asserted label (CWE-200). - Operationalize context minimization with an explicit include/exclude allowlist; restrict external-advisor subagents to read-only/no-Bash/ no-filesystem scope (CWE-269). - Add MCP-router availability pre-flight, a security trigger keyword, and minor wording fixes. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Address review findings on the dev-prohibition seeder: - Ship the rule definitions in-repo at scripts/memory/seed-data/ dev-prohibitions.json (the 4 canonical rules, no secrets) and default SEED_FILE to it, with SEED_FILE=<path> override. The seeder no longer depends on an out-of-repo ~/.ralph file, so it works on a clean clone. - Reject unknown arguments with FATAL/exit 1 instead of silently defaulting to the mutating --apply path; validate args before prerequisite checks (fail fast on misuse). - Stop suppressing recall_v2 stderr so infra failures surface their real root cause instead of a misleading "node not surfaced". - Add tests/memory/test-seed-dev-prohibitions.sh: fail-loud contract tests (missing seed -> exit 1, unknown arg -> exit 1, fixture valid). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Nine skills shipped their definition as lowercase skill.md, which is indistinguishable from SKILL.md on macOS (case-insensitive) but breaks the 'Validate skills structure' CI step on Linux (case-sensitive), where the check looks for an exact SKILL.md. Rename all nine to the canonical SKILL.md so the gate passes on a clean clone. Affected: attack-mutator, context7-usage, defense-profiler, gemini-cli, minimax-mcp-usage, tap-explorer, task-visualizer, vercel-react-best-practices, worktree-pr. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…le refs AutoReview (passes 4-5) found two real gaps, both consequences of this PR: - The new seeder contract test (test-seed-dev-prohibitions.sh) was orphaned: pytest only collects test_*.py so CI never ran it. Add a pytest wrapper (tests/memory/test_seed_dev_prohibitions.py) that runs the bash contract test under `pytest tests/`, failing loudly on any broken contract. - The 9 skill.md -> SKILL.md renames left two committed tests asserting the old lowercase path; update test_v2.25_search_hierarchy.sh and test_v2.26_prefix_commands.sh to SKILL.md. - Gitignore .claude/tasks.json (ephemeral task-tracker state). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
alfredolopez80
added a commit
that referenced
this pull request
Jun 22, 2026
…regression fix(skills): restore Aristotle Integration regression from #26
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds
scripts/memory/seed-dev-prohibitions.sh— a one-command, idempotent wrapper that graduates the 4 global development-prohibition rules into the project-scoped Ralph Memory Tree so they surface inrecall_v2(Codex recall).The 4 rules:
What the script does
migrate_rules_to_nodes.py --rules ~/.ralph/procedural/seed-dev-prohibitions.json --apply→ creates GREEN nodes (confidence 1.0).project_memory.py --apply→ refreshes the read-only GREEN projection inMEMORY.md.recall_v2verification → fails loud if any of the 4 nodes is not surfaced.The script practices the rules it seeds:
set -euo pipefail,umask 077, explicit fail-loud prerequisite checks, no silent fallbacks.Design notes
updated: 4, created: 0).recall_v2hard-rejects foreignproject_ids (intentional per-project isolation). Run per-repo where in-repo Codex recall is wanted.~/.claude/CLAUDE.mdand~/.claude/rules/proven/*.md(outside this repo). This PR only adds the in-repo recall tooling.Scope
Only
scripts/memory/seed-dev-prohibitions.shis repo-tracked. The CLAUDE.md directive, proven rule files, the seed JSON, and the Memory Tree nodes all live outside the repo (~/.claude/,~/.ralph/) and are not part of this PR.🤖 Generated with Claude Code