why
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize external, user-controlled texts such as PR bodies, code comments, issue trackers, commit messages, and chat histories, creating a potential surface for indirect prompt injection.
- Ingestion points: Untrusted text enters the agent context via
SKILL.md,references/anchor.md, and specific playbooks withinreferences/sources/through commands likegit log,gh pr view, and custom MCP server tools. - Boundary markers: The instructions ask for verbatim quotes and full content readings without enforcing strict delimiters or structured formatting to distinguish data content from instructions.
- Capability inventory: Capabilities are primarily read-only data aggregation commands including
git blame,git log,gh pr view, andrg(ripgrep); no dangerous file-writing or arbitrary command execution capabilities are present. - Sanitization: Explicit escaping, sanitization filters, or validation mechanisms for the fetched text inputs are absent.
Audit Metadata