Enterprise Retro Memory (governed)

ARGUMENTS: $ARGUMENTS

Mode

• full (default): evidence-ledger + update all .claude/rules/*.md + run validate_rules.sh
• light: evidence-ledger + update only active-context.md + run validate_rules.sh

Guardrails (anti-hallucination)

• Use ONLY information from the current conversation + existing repo files.
• Anything not explicit must be marked as Assumption (confidence: high/med/low), or excluded.
• Any inferred rule MUST include "Evidence:" referencing what in the chat supports it.
• Never store secrets/tokens/credentials. If found, redact and add "[REDACTED]".

Step A — Create Evidence Ledger (main context)

1. Ensure directories exist:

   • ./.claude/retro/

2. Write a new file: ./.claude/retro/-evidence.md Use format: YYYY-MM-DDTHH-MM (e.g. 2026-02-17T14-30-evidence.md)

   Include sections:

   • Decisions & Agreements (explicit only)

   • Preferences (explicit only)

   • Failure Modes Observed (repeated problems)

   • Inefficient Loops to Avoid

   • Effective Patterns to Repeat

   • Open Questions / Unknowns

   • Assumptions (minimize; include confidence)

   • Redactions (if any)

3. Keep it short but evidence-rich. Use bullet points. Short quoted fragments (<= 1 line each).

Step B — Delegate to isolated subagent

Use the retro-auditor agent and pass:

• ledger path
• mode

Prompt: "Use the retro-auditor agent to process ./.claude/retro/-evidence.md in MODE=<full|light> and update ./.claude/rules accordingly."

Step C — Validate & Fix

Run:

bash .claude/skills/retro-memory-enterprise/scripts/validate_rules.sh

If warnings/errors occur, fix and re-run until clean.

Step D — Return summary

Output in chat:

• Ledger path created
• Files updated (list)
• Top 5 rule changes
• Assumptions introduced (if any)
• Contradictions resolved (if any)