redact

# /redact. Memory audit primitive Strip sensitive content from any tracked file with an audit trail. Pattern modeled on the managed-agent memory feature (filesystem audit + per-entry redaction), adapted to a solo-operator stack. ## When to use - A credential, API key, or token slipped into a memory file or output - A name, phone number, or address you didn't mean to log durably - A quote from a private conversation that shouldn't sit in a corpus that may go public (e.g. for a public review or backups) - Any case where you want a clean strip + audit trail (NOT silent deletion) ## When NOT to use - Removing entire memory files → use `git rm` + commit - Bulk cleanup → use the `dreaming` skill's pruning pass instead - Nothing got tracked yet → just don't write it ## Workflow ### Invocation ``` /redact <file-path> "<search-pattern>" "<reason>" ``` Examples: ``` /redact memory/Context/partner-event.md "sk-XXXX..." "leaked API key" /redact outputs/raw/agent/2026-04-25-self-audit.md "<counterparty name>" "PII. anonymize counterparty" /redact memory/Decisions.md "+1555..." "phone number leaked" ``` ### Steps the skill executes 1. **Read the target file.** Verify it exists and is in `memory/` or `outputs/` (refuse on `.env`, `.ssh/`, anything outside the workspace). 2. **Find matches.** Show every line containing the pattern (with line numbers). User confirms before any change. 3. **Replace each match in place** with `[REDACTED YYYY-MM-DD reason="<reason>"]` (pattern length ...