subagent-review-cycles
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to audit and sweep content produced by other agents or users, which may contain adversarial instructions designed to influence the agent's logic.
- Ingestion points: The Code Reviewer subagent reads code or documents generated by the Developer as defined in SKILL.md. The Leader runs a grep/ripgrep sweep over file contents as defined in references/phase-todo-sweep.md.
- Boundary markers: No explicit boundary markers or instructions to ignore embedded prompts are specified.
- Capability inventory: The skill utilizes search tools (rg, grep) and Git commands (git log, git show, git diff).
- Sanitization: No data sanitization or validation of the input content is mentioned.
- [COMMAND_EXECUTION]: The skill provides templates for shell commands to be executed by the agent for searching and diffing files.
- Evidence: Specific commands using git diff, xargs, rg, and grep are provided in references/phase-todo-sweep.md.
- [EXTERNAL_DOWNLOADS]: The skill references documentation from external sources to support its claims.
- Evidence: Links to official documentation for Claude Code and Git are provided in references/official-docs.md.
Audit Metadata