agentic-peer-review
Pass
Audited by Gen Agent Trust Hub on Sep 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of code diffs, pull requests, and integration deltas, which creates a surface for indirect prompt injection.
- Ingestion points: Reviewer agents ingest code from pull requests and branches as defined in the Shared cohort block in references/reviewer-prompts.md.
- Boundary markers: The workflow uses a specific Required output schema and domain-specific briefs to constrain agent behavior.
- Capability inventory: The skill performs command execution for reproductions and deployment verification as described in SKILL.md.
- Sanitization: Safety rules in SKILL.md explicitly mandate the use of fake data, fake secrets, and prohibit copying production data into records or prompts.
- [COMMAND_EXECUTION]: The workflow involves executing commands to run reproductions, promote tests to the product suite, and verify deployments.
- Evidence: Step 6 (reproduce before ranking) and Step 8 (fix in priority order) involve running tests. Step 10 (verify the deployment) includes running unauthenticated probes and checking entrypoints.
- [EXTERNAL_DOWNLOADS]: The skill includes instructions to install itself using npx.
- Evidence: README.md provides the command npx skills@latest add looties-io/looties-skills --skill agentic-peer-review.
Audit Metadata