The Agent Skills Directory

[SAFE]: The skill provides stylistic and linguistic constraints for the AI agent to follow when generating human-facing text, such as PR descriptions and status updates. It focuses on clarity and the removal of technical jargon.
[NO_CODE]: The skill consists exclusively of Markdown documentation and JSON-formatted test cases. No executable code or scripts are included in the package.
[INDIRECT_PROMPT_INJECTION]: The skill instructions define an 'outer register' for human communication and provide a 'Sweep list' to strip technical tokens from processed data. This acts as a defensive measure to prevent internal session data from leaking into public-facing documentation. The mandatory evidence chain for this attack surface is: ingestion occurs from session notes/ledgers; boundary markers are established between inner and outer registers; capabilities are limited to writing standard repository documentation (PR.md, REPLY.md); and sanitization is implemented via the explicit sweep list of forbidden tokens.

speak-outer