retro-run
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub pull requests, including review comments and PR descriptions, to generate triage reports. This creates an attack surface where a malicious PR could contain instructions aimed at manipulating the agent's analysis or causing it to file deceptive issues.
- Ingestion points: Data is ingested via
scripts/retro.sh candidateswhich fetches PR bodies and comments from the GitHub API. - Boundary markers: The instructions do not specify any delimiters or safety warnings for the agent when processing this external content.
- Capability inventory: The skill has the ability to create new issues in the repository using the
ghCLI viascripts/retro.sh file. - Sanitization: The bash script performs basic validation on the
findings.jsonschema (e.g., kebab-case keys, single-line summaries) and usesjqfor safe JSON construction, but it does not sanitize the natural language content against prompt injection. - [COMMAND_EXECUTION]: The skill relies on the execution of a local shell script
scripts/retro.shto interact with the GitHub API. While the script is part of the skill package, it represents a direct command execution capability that interacts with the repository's host environment.
Audit Metadata