improve-pipeline
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external projects (field incidents) to diagnose pipeline weaknesses. This external input is used to drive the reasoning process and generate GitHub issues. This represents a potential attack surface where malicious data in a project could influence the meta-skill's output. The risk is mitigated by mandatory human review (Phase 6) and explicit targeting of the author's own repository for all automated writes.
- Ingestion points: Incident context from downstream project workspaces (SKILL.md Phase 1).
- Boundary markers: Explicit Phase 6 review gate and the requirement for human approval for Phase 7 implementation.
- Capability inventory: GitHub CLI (gh) issue and label management, local library tool access, and optional file implementation capabilities.
- Sanitization: The skill enforces strict structured formatting for all output (referenced in writing-for-humans.md).
- [COMMAND_EXECUTION]: The skill executes shell commands using the GitHub CLI ('gh') to list and create issues and labels. These commands are hardcoded to target 'chrislacey89/skills', ensuring that the impact is limited to the author's controlled environment. It also utilizes a local '/library' tool to query a book index at a known path.
Audit Metadata