enforcee-auditlisted

Audit whether an AI output followed a ruleset, rule by rule, with evidence. Use when the user asks whether their rules were followed, whether CLAUDE.md is being honoured, why the assistant ignored an instruction, to check compliance of a response against project rules, or to produce a receipt for a piece of work. Also use when the user wants to know which of their rules are dead, contradictory or too vague to check.
7patriksz-art/Enforcee · ★ 0 · AI & Automation · score 53

Install: claude install-skill 7patriksz-art/Enforcee

# Auditing rules with Enforcee Enforcee answers one question: **which of these rules did the output actually follow, and where is the proof?** ## When to reach for it - "Did you follow my CLAUDE.md?" - "Why do you keep ignoring rule X?" - "Check this answer against our style guide." - "Which of my rules are actually doing anything?" ## How to run it Everything runs locally and makes no network call unless `--judge` is passed. ```bash # Audit an output against a ruleset. Exits non-zero if any rule was VIOLATED. npx enforcee audit CLAUDE.md answer.md # Critique the ruleset on its own — duplicates, contradictions, unenforceable rules. npx enforcee health CLAUDE.md # Propose rules from things the user already said in conversation. npx enforcee learn conversation.txt CLAUDE.md # Read a session transcript: skills offered vs used, MCP servers that never connected. npx enforcee session ~/.claude/projects/<project>/<session>.jsonl ``` ## Reading the output Each rule gets one of four verdicts, and the method that produced it: - `FOLLOWED` / `VIOLATED` — backed by an exact quote at a character offset. - `NOT_APPLICABLE` — the rule's trigger never fired. - `UNVERIFIABLE` — no observable signal either way. This is a respected answer, not a failure. `proof` means code decided it and the result is reproducible. `judge` means a model decided it, and its evidence quote was verified to exist literally in the output or the verdict was discarded. **Coverage** is the headline number