claude-md-improver

Pass

Audited by Gen Agent Trust Hub on Jul 11, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill provides a structured framework for managing repository-level guidance files (CLAUDE.md) and navigation maps (code_map.md), which are standard components of the Claude Code development ecosystem.
  • [SAFE]: The instructions include robust safety boundaries, specifically directing the agent to never read the contents of personal developer overrides (CLAUDE.local.md) and to never attempt edits on managed OS policies.
  • [COMMAND_EXECUTION]: The skill uses a restricted set of shell commands (Bash(find *) and Bash(git *)) for repository discovery and verification of existing project structures, which is appropriate for its stated purpose as a developer tool.
  • [DATA_EXFILTRATION]: There are no network-capable tools allowed or used within the skill. All analysis and file manipulation occur within the local project environment, precluding data exfiltration risks.
  • [INDIRECT_PROMPT_INJECTION]: Although the skill processes untrusted markdown content from repository files, it mitigates manipulation risks by enforcing a rigorous quality rubric and multiple verification phases (Discovery, Evidence Collection, and Final Verification) that require the agent to ground its changes in verifiable repository artifacts.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 11, 2026, 04:49 PM
Security Audit — agent-trust-hub — claude-md-improver