The Agent Skills Directory

[SAFE]: The skill exhibits no malicious behaviors and incorporates proactive security measures to prevent the accidental disclosure of sensitive information.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data (source code, diffs, and patches), which constitutes an indirect prompt injection surface.
Ingestion points: Untrusted content is read during the teaching workflows described in SKILL.md (Workflow steps 1 and 2), guides/changes.md, and guides/whole-application.md.
Boundary markers: While the skill lacks explicit markers to ignore instructions embedded within the analyzed code, it provides detailed instructions on identifying and excluding sensitive content.
Capability inventory: The skill is restricted to read-only analysis and explanatory output, with no detected network or file-write capabilities.
Sanitization: The skill explicitly requires the redaction of secrets, API keys, and private personal data before quoting or summarizing any external content, as detailed in SKILL.md and guides/changes.md.

teach