self-review

Review whether rules were followed in this session.

Instructions

1. Read all rule sources fresh from disk (do not rely on memory). Walk both the shared tier and the current tool's tier:

   ~/.config/ai/PERSONAL.md
   ~/.config/ai/guardrails.md
   ~/.config/ai/rules/*.md
   ~/.config/{{TOOL_NAME_LC}}/DEVGUARD.md      (if exists — Claude carries a thin tool-specific addendum)
   ~/.config/{{TOOL_NAME_LC}}/rules/*.md       (e.g. claude-subagent-trust.md)
   ~/.config/ai/memory/*.md                    (feedback entries only)
   ~/.config/{{TOOL_NAME_LC}}/memory/*.md      (feedback entries only — claude-* / codex-* prefix)
   ~/.config/ai/memory/private/*.md            (feedback entries only)
   

2. Derive the rows from what you read: one row per top-level rule (each ## section, or each top-level bullet in a file without sections) and one row per memory feedback entry, named by its name field. Do not carry a fixed list — a rule that was added, removed, or renamed on disk must show up that way in the review.

3. Evaluate each row against the conversation history:

   • ✓ followed (cite evidence)
   • ✗ violated (cite the specific action/message)
   • — not triggered (explain why) OR — (n/a — rule not present in this tool) if the rule lives in a tier that does not apply to the current tool.

4. Output in this exact format: one ### section per rule source in the walk order above, each a three-column table. Example for guardrails.md:

   ## Self-Review: Rule Compliance
   
   ### guardrails.md (shared)
   
   | Rule | Result | Evidence |
   |------|--------|----------|
   | Verification Before Completion (evidence in the same message) | ✓/✗ | ... |
   | Absence Proofs (scope match + command and output) | ✓/✗/— | ... |
   | Scope Resolution | ✓/✗/— | ... |
   | Restored Context Authority (summary-only shorthand) | ✓/✗/— | ... |
   | Pre-Implementation Gate (leanness) | ✓/✗/— | ... |
   

5. Close with the tally and, only when at least one row is ✗, the actionability table:

   ### Actionability
   
   **Followed**: N / **Violated**: N / **Not triggered**: N
   
   | Violation | Fix | Feasibility |
   |-----------|-----|-------------|
   | ... | hook / rule change / code change | 즉시 가능 / 규모별 분기 / 불가 (구조적 한계 — reason) |
   

Be brutally honest. Do not soften or omit violations.