The Agent Skills Directory

[SAFE]: The skill processes untrusted external specifications and codebases (ingestion points in spec-document and codebase-path). It implements effective boundary markers and sanitization through a structured workflow that mandates literal quotes and line-number citations for all claims, preventing the agent from being influenced by embedded instructions in the audited data.
[SAFE]: The capability inventory (including Read, Grep, Bash, and WebFetch) is appropriately scoped for auditing tasks. The instructions include strict anti-hallucination requirements and reject rationalizations that might lead to speculative or undocumented behavior analysis.
[SAFE]: No malicious patterns such as hardcoded credentials, data exfiltration, obfuscation, or persistence mechanisms were detected in the skill instructions or associated resource files.

spec-to-code-compliance