The Agent Skills Directory

[SAFE]: The skill defines a controlled process for project lifecycle management. It focuses on verifying the presence and status of project artifacts (e.g., PRDs, test results, security checklists) and updating internal project state files. No evidence of malicious behavior, credential theft, or unauthorized network access was found.
[INDIRECT_PROMPT_INJECTION]: The skill evaluates untrusted data from project files, creating a potential surface for indirect prompt injection. However, this is mitigated by human approval requirements and explicit guardrails against social engineering.
Ingestion points: Reads gate criteria verification from PRD.md, PLAN.md, project/TASKS.md, and templates/security/SECURITY_CHECKLIST.md (SKILL.md).
Boundary markers: Uses markdown headers and structured checklists to delimit criteria.
Capability inventory: Performs file writes to agents/memory/episodic/SESSION_SNAPSHOT/, PROJECT_STATE.md, and AUDIT_LOG.md. Executes git tag for versioning (SKILL.md).
Sanitization: Relies on mandatory human-in-the-loop approval for critical phases (S00, S01, S02, S07, S08) to verify agent assertions and prevent social engineering bypasses.

sdlc-checkpoint