The Agent Skills Directory

[SAFE]: No critical, high, or medium security risks were identified. The skill correctly implements its intended logic using standard platform tools and follows a read-only analysis pattern prior to modification.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by processing untrusted data from phase artifacts.
Ingestion points: Reads .workflow/phases/ (index.json, plan.json, verification.json, review.json, uat.md, and task summaries).
Boundary markers: Absent. The lens agents receive artifact data directly via template strings.
Capability inventory: Filesystem writes via apply_patch, issue creation in issues.jsonl, and memory capture through the manage-memory-capture skill.
Sanitization: No sanitization or escaping of artifact content is documented.
Assessment: This exposure is intrinsic to the skill's primary function (retrospective analysis) and is assessed as safe given the context of project history processing.

quality-retrospective