The Agent Skills Directory

[SAFE]: No malicious code, command execution, or data exfiltration patterns were found. The skill is designed purely for product management and stage-gate reviews.
[PROMPT_INJECTION]: No evidence of prompt injection attempts aimed at overriding system guidelines. The skill explicitly instructs the agent to treat artifacts (decks, repos) as untrusted evidence rather than instructions, which serves as a defense against injection within processed data.
[DATA_EXFILTRATION]: No sensitive file access (e.g., .ssh, .aws) or network exfiltration patterns were identified. References to external URLs are for documentation purposes and point to legitimate tech-focused sites.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input data from project artifacts but implements a robust defense-in-depth posture.
Ingestion points: Gate decks, repositories, logs, and telemetry files specified in SKILL.md.
Boundary markers: Explicit 'Review Posture' in SKILL.md instructing the agent to ignore any directives found within reviewed artifacts.
Capability inventory: The skill does not possess capabilities to write to the filesystem, execute subprocesses, or perform network requests.
Sanitization: Content is processed as evidence to be graded based on documentation rather than as instructions to be followed.
[DYNAMIC_CONTEXT_INJECTION]: No use of the exclamation-backtick syntax (!command) was found in the SKILL.md file.

ai-stage-gate