The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands for environment setup, including mkdir for temporary scratch directories and python3 to run internal validation scripts like validate-frontmatter.py and validate-doc-claims.py.
[COMMAND_EXECUTION]: It interfaces with git and gh (GitHub CLI) to perform non-mutating checks such as verifying PR status and file reachability within the repository history.
[INDIRECT_PROMPT_INJECTION]: The skill processes user conversation history and codebase content to generate durable documentation, creating a surface for indirect prompt injection.
Ingestion points: Conversation history and codebase files read during the research phase.
Boundary markers: None explicitly defined to encapsulate interpolated user data.
Capability inventory: Capability to write to project files (docs/solutions/, CONCEPTS.md) and execute shell commands (git, gh, python3).
Sanitization: Employs validate-frontmatter.py for YAML parser-safety and validate-doc-claims.py to mechanically verify that cited paths and SHAs exist.
[EXTERNAL_DOWNLOADS]: The skill utilizes git fetch and gh pr view to retrieve metadata from GitHub. These operations target well-known services and are used exclusively for verification purposes.
[DYNAMIC_EXECUTION]: The bundled script validate-doc-claims.py uses subprocess.run to execute git commands. This usage is implemented using best practices (argument lists instead of shell strings) and is scoped to standard repository operations.

capture-learning