audit

Reconstructs decision history from logs and primary sources to analyze context gaps and pin corrective knowledge.

3|1|Updated Dec 9, 2025
One-click install
npx skills add https://github.com/isizono/calm --skill audit-isizono
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: audit
Source: https://github.com/isizono/calm/tree/main/skills/audit
Command: npx skills add https://github.com/isizono/calm --skill audit-isizono

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? In long-running projects, past decisions get revisited repeatedly, contradict each other, or are silently overturned, causing circular discussions. This Skill audits the validity of past decisions by reading primary resources and chronological logs, then pins the resulting knowledge to the right place so the same debate never repeats. ## Core Features & Use Cases - Trigger-based activation: Fires on self-detected signals (e.g., a third policy change under the same tag, a bug seen twice, a previously rejected proposal resurfacing) or user phrases like "didn't we already decide this?" - Seven-step audit process: Documents the trigger, fixes scope to one subject, fetches decisions and supersedes chains, reads related logs chronologically, maps the evolution of the issue, and analyzes structural causes of context gaps. - Knowledge pinning matrix: Routes findings to the correct destination (tag notes, habits, anchors, materials, or decision revision proposals) with clear rules for autonomous execution versus user approval. - Use Case: When you notice the same design topic being debated for the third time, the audit reconstructs the full decision timeline, identifies why the judgment drifted, and records an audit material with a completion marker to prevent duplicate audits within 24 hours. ## Quick Start Tell the AI "this feels like something we decided before — audit the past decisions on this topic" and it will reconstruct the decision history and propose where to pin the corrective knowledge.

Frequently Asked Questions about audit

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I audit past decisions in a long-running project?▼

Trigger the audit when you suspect a past decision is invalid or discussions are looping. It fetches the decision, its supersedes chain, anchors, and up to 30 recent logs within 90 days, then builds a timeline map and analyzes why the judgment drifted.

What triggers a decision audit automatically?▼

Five self-detected triggers fire it: doubting a decision right after citing it, a third policy change under one tag, sensing a topic was already discussed, seeing the same bug twice, or recognizing a previously rejected proposal. User phrases like "are we going in circles?" also trigger it.

Can the audit retract or supersede a decision directly?▼

No. Retracting or superseding decisions is outside the audit scope. The audit records revision proposals in the audit material's verification results table, and the actual decision change always goes through user adjudication.

How does the audit prevent duplicate audits on the same topic?▼

Before fixing scope, it searches for audit materials on the same subject within 24 hours and checks for an #audited-YYYY-MM-DD marker in the representative tag note. If a recent audit exists, it references the prior material instead of running a new one.

Where does audit knowledge get pinned after analysis?▼

A decision matrix routes knowledge by type: tag-specific rules go to tag notes, session-wide rules to habits, truth-verification targets to anchors, and investigation results to materials. Pure additions run autonomously while rewrites and deletions require user confirmation.

What happens if an audit is interrupted mid-session?▼

The partial audit material is saved as a draft with a status line noting the step where it stopped. On resumption, the same material is updated rather than creating a new one, preserving the interruption point.