The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates and engineering patterns for workflows that process untrusted external data, which creates a surface area for potential indirect prompt injection attacks.
Ingestion points: The guide (SKILL.md) provides instructions for workflows triggered by on: issues, on: pull_request, and on: issue_comment events, all of which contain data provided by external users.
Boundary markers: The prompt examples provided in the markdown body use natural language and markdown headers but do not demonstrate the use of explicit boundary delimiters (e.g., XML tags) to isolate untrusted input from system instructions.
Capability inventory: The workflows documented include tools with significant write capabilities, such as safeoutputs___file, safeoutputs___comment, and safeoutputs___label.
Sanitization: The authoring instructions focus on instruction clarity but do not include guidance on sanitizing or validating external input before it is processed by the AI engine.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing necessary compiler tools from an official package registry.
Evidence: The documentation in SKILL.md includes instructions to install the @github/agentic-workflows-compiler package using the npm package manager.

gh-aw-workflow-authoring