claude-code-audit
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes several internal Python scripts (
inventory.py,sample.py,extract.py,render.py) located within the skill's own directory to process local session data. - [DATA_EXPOSURE]: The skill accesses local Claude Code session logs located in
~/.claude/projects/and configuration files like~/.claude/CLAUDE.md. This is the primary intended function of the skill (auditing sessions) and the data remains local to the user's environment for analysis. - [DYNAMIC_CONTEXT_INJECTION]: While the skill is designed for the Claude Code platform, it does not use the
!command`` syntax in itsSKILL.mdfor pre-execution, nor does it inject untrusted user input into shell commands. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (past session transcripts). However, it implements a
render.pyscript that clips and sanitizes the output before passing it to subagents, reducing the risk of embedded instructions influencing the auditor. The severity is low as this is a diagnostic surface. - [TRUST_SCOPE_RULE]: The skill relies on standard Python libraries and does not download external dependencies or remote code during execution.
Audit Metadata