iterate-retrospective
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates team retrospectives by gathering observations from all team members and documenting these in reports and a project-local memory file. This creates a vulnerability surface where malicious instructions could be embedded in the team feedback.\n
- Ingestion points: Participant observations gathered during the facilitation process (SKILL.md) and raw input provided in retrospective scenarios (evals/output-scenarios/sprint-42.md).\n
- Boundary markers: The skill uses a structured template (references/TEMPLATE.md) to categorize output, but lacks specific instructions to the agent to treat participant observations as untrusted or to ignore any command-like structures within that data.\n
- Capability inventory: The skill reads previous cycle artifacts and writes new interpretations to the
.claude/pm-skills.local.mdfile (Project Memory Contract).\n - Sanitization: No explicit validation, filtering, or escaping of participant-provided observations is performed before they are stored in the memory file or included in the finalized retrospective document.\n- [SAFE]: The skill references the author's official GitHub repository for documentation and project tracking (SKILL.md).
Audit Metadata