The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines routines that ingest untrusted content from external sources, specifically GitHub issue bodies and pull request review comments. This creates a surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through gh pr view, gh issue view, and GraphQL API calls to fetch review comments.
Boundary markers: The routines/pr-feedback-resolve.md file includes an explicit warning to treat review comment text as untrusted and to never execute code or URLs from comments.
Capability inventory: The skill has the capability to write to the repository (git commit/push) and interact with GitHub (gh issue/pr comment).
Sanitization: No programmatic sanitization is performed; the skill relies on instructional guardrails to prevent exploitation.
[COMMAND_EXECUTION]: The skill instructions utilize common CLI tools including git and gh (GitHub CLI) for repository management. It also provides for the execution of project-specific scripts (e.g., scripts/verify-install.sh and bun run scripts) as part of health checks. These are standard operations for the skill's intended purpose of codebase automation.
[EXTERNAL_DOWNLOADS]: The REFERENCE.md file contains examples for triggering routines via Anthropic's official API endpoints. These references target a well-known service and are used appropriately for integration.

setup-routines