The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides the sh and tsx verbs to execute arbitrary shell commands and TypeScript code within temporary, isolated 'world' directories. It also interacts with the gh (GitHub) CLI, which may perform actions on real repositories if configured by the user.
[DYNAMIC_EXECUTION]: Uses the tsx executor to run TypeScript scripts that are dynamically created or modified within the verification environment.
[EXTERNAL_DOWNLOADS]: Fetches peer dependencies from well-known providers during setup, including packages from Vercel, Daytona, E2B, Modal, and Fly.io. It also handles the download of binaries for the 'beads' issue tracker.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external issue trackers (GitHub, Beads) which could contain malicious instructions for the agent.
Ingestion points: Untrusted issue text is retrieved via issue tracker integration files.
Boundary markers: The skill uses isolated 'world' directories and requires human-initiated commands for specific runs.
Capability inventory: Includes shell execution (sh), TypeScript execution (tsx), filesystem operations, and git commands.
Sanitization: Explicitly implements pattern matching to detect and report shell injection payloads (e.g., subshell execution) found in external issue text.

verify