The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates pull request creation by executing standard command-line tools including git and platform-specific CLIs like gh and glab. Evidence: Phases 1, 2, 3, 6, and 8 utilize shell commands for repository inspection, diff analysis, and PR submission. Safety: The skill enforces strict user-in-the-loop confirmation gates (Phases 2.2, 3.2, and 8.1) before any command that alters the repository state or creates external records is executed.
[PROMPT_INJECTION]: The skill processes untrusted repository content, such as diffs and CODEOWNERS files, creating a surface for potential indirect prompt injection. Ingestion points: Phase 3.3 (loading full git diff) and Phase 6.1 (reading CODEOWNERS). Boundary markers: No explicit boundary markers or instructions to treat the ingested data as untrusted are included in the prompt construction. Capability inventory: The agent has access to file system read operations and shell command execution. Sanitization: No sanitization or validation of the repository data is performed prior to the agent's interpretation.

pr-creator