The Agent Skills Directory

[SAFE]: The skill instructions and associated reference papers promote best practices for production safety, including concurrency testing, failure injection, and resource management. No malicious patterns, data exfiltration, or obfuscation were detected.
[INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted code and artifacts, creating a potential surface for indirect prompt injection. However, the skill provides extensive mitigations through its evidence-state framework and boundary closure requirements.
Ingestion points: Untrusted repositories, code diffs, and deployment artifacts are ingested during the Review and Verify modes in SKILL.md.
Boundary markers: The skill uses a strict Release Decision Gate and Companion Skill system in SKILL.md to define authority and scope.
Capability inventory: The Verify mode instructions in SKILL.md imply the execution of tests and diagnostic tools on the local environment.
Sanitization: The Release Decision Gate enforces a requirement that all evidence must be executed and inspected, specifically forbidding the conversion of planned or claimed states into valid proof.

quality-release