The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for building agents that ingest untrusted data from GitHub Issues, Pull Requests, and external web sources. This creates a surface for indirect prompt injection attacks.
Ingestion points: Untrusted data enters the agent context through GitHub event triggers (on: issues, on: pull_request) and the web-fetch tool as described in SKILL.md.
Boundary markers: The documentation specifically recommends the use of safe-inputs with strict schemas and markdown delimiters to isolate untrusted data.
Capability inventory: The skill describes tools for GitHub API access (github:), shell command execution (bash:), and file system modification (edit:, create:).
Sanitization: Security mechanisms like "Threat Detection" and "Safe Outputs" (which limit write operations) are documented as primary sanitization and risk mitigation strategies.
[EXTERNAL_DOWNLOADS]: The documentation references external resources and extensions from the official GitHub organization (github/gh-aw) and the vendor's own repository (Hack23/riksdagsmonitor). These references point to trusted or author-owned sources.
[SAFE]: Security best practices are followed throughout the documentation. For example, all mentioned credentials use standard GitHub Actions secret syntax (e.g., ${{ secrets.GITHUB_TOKEN }}) rather than hardcoded values.

github-agentic-workflows