auditing-allowed-tools

Audits and patches allowed-tools declarations in SKILL.md files against session transcripts.

Updated Jun 8, 2026
One-click install
npx skills add https://github.com/shawn-sandy/agentics-kit --skill auditing-allowed-tools-shawn-sandy
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: auditing-allowed-tools
Source: https://github.com/shawn-sandy/agentics-kit/tree/main/kit/plugins/skill-reviewer/skills/auditing-allowed-tools
Command: npx skills add https://github.com/shawn-sandy/agentics-kit --skill auditing-allowed-tools-shawn-sandy

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? SKILL.md files often declare the wrong allowed-tools set — either missing tools that trigger mid-run permission prompts, or declaring overly broad permissions that were never needed. Manually reconciling frontmatter declarations against what a skill actually invokes is tedious and error-prone. ## Core Features & Use Cases - Static audit: Scans a SKILL.md body for tool evidence (named tools, Bash CLI families, MCP references) and diffs it against the declared allowed-tools, classifying each entry as OK, MISSING, or UNUSED. - Session audit: Parses a Claude Code session JSONL transcript via a bundled Python script to report which tools were actually invoked, including per-command Bash breakdowns. - Cross-reference mode: Compares declared, statically detected, and observed tool sets to surface undocumented runtime dependencies and clear removal candidates. - Use Case: After writing a new skill, ask to audit its allowed-tools; the skill recommends a minimal permission set and can patch the frontmatter line directly after your confirmation. ## Quick Start Ask the assistant to audit the allowed-tools declaration for a specific SKILL.md path and apply the recommended fixes.

Frequently Asked Questions about auditing-allowed-tools

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

FAQPage Schema
How do I audit allowed-tools in a SKILL.md file?▼

Provide the path to the SKILL.md and ask for an allowed-tools audit. The skill scans the body for tool evidence, diffs it against the declared frontmatter line, and reports each tool as OK, MISSING, or UNUSED before offering to patch the file.

How to check which tools Claude actually used in a session?▼

Run a session audit by providing a session UUID or JSONL path, or ask about the current session. The bundled session_tool_scan.py script parses the transcript and reports per-tool call counts plus a per-command Bash breakdown.

Can the skill automatically fix missing allowed-tools entries?▼

Yes. After the audit report, it offers three options: add only missing tools, replace with the minimal recommended set, or report only. Edits touch only the allowed-tools line and are confirmed by re-reading the file.

Does the audit handle MCP tool references?▼

MCP tools matching the mcp__server__tool pattern are detected and reported as informational notes, but they are never written into the allowed-tools frontmatter because the declaration format for MCP tools is unconfirmed.

What are the limitations of allowed-tools auditing?▼

The skill only audits SKILL.md allowed-tools declarations; it does not touch settings.json permission rules, hooks, or command/agent frontmatter. It also does not score general SKILL.md quality — that belongs to a separate reviewing skill.