agent audit

Security Community

Skill notes

Gotchas

Flagging intentional patterns as issues. Read CLAUDE.md and .claude/rules/ FIRST. What looks like a misconfiguration may be an intentional project convention documented there. Never report something as an issue if it's explained in the project's own docs. Recommending fields that don't exist. Claude Code's supported frontmatter fields change between versions. ALWAYS verify against the latest docs (Phase 1 research) before flagging a "missing" field — it may not be supported yet, or may have been deprecated.

Argument Parsing

Parse $ARGUMENTS to determine scope. Examples: Invocation · Scope /agent-audit · Full audit — all components /agent-audit full · Same as above /agent-audit hooks · Only hooks (settings.json hook config + .claude/hooks/ scripts)

Phase 1: Research Latest Best Practices

Use the Agent tool to launch parallel research agents (subagent_type: "general-purpose"). All agents MUST search the web for the current year's content:

Agent 1: Official Claude Code Docs

Fetch latest official Claude Code documentation for: Hooks: events, matchers, exit codes Settings: permissions, deny lists, sandbox, attribution Memory: CLAUDE.md hierarchy, auto-memory, .claude/rules/ Skills: valid frontmatter fields (critical — verify which fields are actually supported vs silently ignored)

Agent 2: Community Patterns & New Features

Search for: Claude Code changelog / release notes for current year Community best practices and patterns New features or deprecated settings When scoped to a component (e.g., /agent-audit hooks): only research that component's docs, not everything.

Fluxly profile Runtime and permissions are inferred by Fluxly static checks, not author declarations or runtime test results.
Domain
Security
Author-declared agents
No explicit declaration found; this is not inferred or tested compatibility
Static check
88 / 100 · heuristic scan, not runtime safety proof
Runtime evidence
Not run; no isolated runtime test evidence is currently recorded
Author / version / license
@tomevault-io · no license declared
Token usage
Lean
Setup complexity
Guided setup
External API key
Not required
Operating systems
Unspecified (assume cross-platform)
Runtime requirements
No explicit requirements detected
Detected file/system behavior
  • Read-only
  • Write / modify
  • Shell exec
  • Env read
Network behavior
External requests
Install commands
None (reference only)

Heads up: Tool permissions are not declared; actual access depends on the runtime.

01
Phase 1: Research Latest Best Practices

Use the Agent tool to launch parallel research agents (subagent_type: "general-purpose"). All agents MUST search the web for the current year's content:

02
Phase 2: General Audit

Read and audit agent-related files. Skip this phase entirely for `skills` or `skill:<name>` arguments.

03
Phase 3: Skill Deep-Dive

Triggered by: `skills` (all), `skill:<name>` (one), or during `full` audit if a skill looks problematic in surface-level check.

04
Phase 3b: Agent Deep-Dive

Triggered by: `agents` (all), `agent:<name>` (one), or during `full` audit.

05
Phase 3c: Rules Conflict Detection

Triggered by: `rules`, or during `full` audit.

06
Phase 4: Report

Phase 4: Report

07
Phase 5: Implementation Plan

If actionable improvements found: Write concrete plan with exact file changes Wait for user approval On approval, implement changes If no issues: report "All clear — configuration is up to date" and skip planning.

Discussion

Powered by GitHub Discussions. Sign in with GitHub to comment, react, or subscribe.