post-review-loop
Pass
Audited by Gen Agent Trust Hub on Jun 17, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill performs shell command execution to run validation tasks (such as test suites) found within the repository's existing workflow. This behavior is standard for development-oriented agents and is limited to local execution.
- [PROMPT_INJECTION]: The skill implements a persistent workflow that relies on an external ledger file (
reviews/post-review-loop.md) for state management. This creates a surface for indirect prompt injection, as the agent is instructed to treat the ledger content as authoritative. A malicious actor with write access to the repository could attempt to influence the agent's logic by modifying the ledger. - Ingestion points: The skill reads the
reviews/post-review-loop.mdstate file, repository source files, and git diff outputs. - Boundary markers: Not present. The agent is instructed to trust the ledger as the primary source of truth for the loop state.
- Capability inventory: The skill possesses the capability to modify repository files and execute arbitrary shell commands for project validation.
- Sanitization: No explicit sanitization or filtering of instructions embedded within the reviewed code or the ledger file is specified.
Audit Metadata