claude-exec
Pass
Audited by Gen Agent Trust Hub on Apr 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides instructions for invoking the Claude Code CLI (claude) via the Bash tool. It details safe patterns for passing prompts through stdin and using flags like --no-session-persistence to manage session data and --max-turns to prevent infinite loops.
- [PROMPT_INJECTION]: The skill facilitates tasks that involve reading untrusted data (codebases and plans) which presents a surface for indirect prompt injection.
- Ingestion points: Untrusted data enters the context via git diff, Read, and Grep operations on the local repository, specifically targeting files like PLAN.md and source code.
- Boundary markers: The skill instructs the agent to use explicit role-playing markers ("You are a strict code reviewer") and safety constraints ("IMPORTANT: Research only
- do not edit any files") in delegated prompts.
- Capability inventory: The delegated sub-agents and CLI calls have access to powerful tools including Bash, Read, Glob, and Grep.
- Sanitization: The skill relies on natural language instructions to constrain the sub-agent's behavior rather than programmatic sanitization of the ingested data.
Audit Metadata