| name | rules-audit |
| description | Use when the user asks to "audit my session", "check for sycophancy", "review behavioral compliance", or otherwise wants a self-discipline pass — scans the current transcript for sycophancy, unnecessary apologies, scope creep, focus drift, and filler language, then reports violations against the rules in `behavioral-core/hooks/rules.d/`. |
| when_to_use | Reach for this near the end of a long session, after a noticeable stretch of low-discipline turns, or when assessing whether the behavioral-anchor hooks are actually shaping output. Do NOT use for real-time rule enforcement — use `/safe-mode` instead; rules-audit only inspects after the fact. |
| disable-model-invocation | true |
| allowed-tools | ["Read","Grep","Glob"] |
| logical | report includes counts for sycophancy, apologies, scope-creep, focus, and filler with an overall discipline rating |
Rules Audit
Review the current conversation for behavioral violations. Scan for and report:
1. Sycophancy Check
Flag instances of: "You're right", "Great question", "That's a great idea", "Absolutely", "Great catch", "Excellent point", "Good thinking", or any reflexive agreement without substance.
2. Unnecessary Apologies
Flag: "Sorry", "I apologize", "My apologies" — unless responding to actual harm caused.
3. Scope Creep
Flag instances where code was added, refactored, or documented beyond what was explicitly requested. Look for:
- Added features not asked for
- Refactored surrounding code during bug fixes
- Added docstrings/comments/types to untouched code
- Over-engineered abstractions for simple tasks
4. Focus Violations
Flag instances of reading files unrelated to the current task, exploring without clear purpose, or switching context without being asked.
5. Filler Language
Flag: "Let me...", "I'll go ahead and...", "Sure, I can help with that!", trailing summaries of completed work.
6. Formatting Defaults
Flag prose that was force-fit into bullet points, numbered lists, or bold headers when plain sentences would do — list scaffolding on a one- or two-item answer, headers on a short reply.
7. Question Restatement & Padding
Flag turns that repeat or paraphrase the user's question before answering, or pad after the point is made — answer length disproportionate to a simple question.
8. Hedged Non-Answers
Flag "which is better, A or B?" turns answered with a symmetric pros/cons list and no pick — the rule requires committing to one and defending it.
Output Format
BEHAVIORAL AUDIT
================
Sycophancy: [count] violations
Apologies: [count] violations
Scope Creep: [count] violations
Focus: [count] violations
Filler: [count] violations
Formatting: [count] violations
Restate/Pad: [count] violations
Hedged Answer: [count] violations
---
Overall Score: [X/10] discipline rating
List each violation with the message where it occurred and what should have been said instead.