The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted merge payloads from external code forges into a database queue for processing by an LLM, creating a potential injection vector.
Ingestion points: Raw webhook payloads are stored in the release_capture_queue table as defined in references/data-model.md.
Boundary markers: The implementation utilizes per-user watermark timestamps and server-side audience filtering as described in references/system-model.md to delimit data exposure.
Capability inventory: The analysis stage orchestrates the describe-changes skill for automated description generation, and the digest job posts these summaries to chat or email surfaces.
Sanitization: The onboarding procedure in SKILL.md explicitly requires that the markdown renderer used for ledger entries has raw HTML passthrough disabled to prevent malicious code injection from commit or PR metadata.
[EXTERNAL_DOWNLOADS]: The skill recommends installing vendor-owned companion skills from the crissmoldovan/agent-skills repository via npx.
[COMMAND_EXECUTION]: The Bash tool is employed to conduct environment discovery and to generate implementation documentation within the host repository.

release-ledger