The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the GitHub CLI (gh) to perform read-only operations such as verifying authentication status, retrieving repository metadata, and viewing issue details. These operations are scoped to the developer's environment and use the tool's built-in authentication.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by ingesting the content of GitHub issues and roadmap entries to populate local documentation.
Ingestion points: Reads external ticket bodies via gh issue view and roadmap descriptions from files within .work/roadmaps/.
Boundary markers: The skill includes explicit instructions to prefix every line of ingested ticket content with quote markers (> ) in the resulting seed.md file, providing a clear boundary between external content and agent instructions.
Capability inventory: The skill has permissions to read project files (AGENTS.md, docs/), write to the .work/threads/ directory, and execute GitHub CLI commands.
Sanitization: Ingested content is reproduced exactly to ensure accuracy for the developer, relying on the structural quoting to mitigate injection risks for downstream processing.

open-thread