The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes git commit logs and diff outputs which may contain untrusted input from external contributors or branch history.
Ingestion points: Ingests untrusted data through git log and git diff commands in SKILL.md (Step 1).
Boundary markers: No explicit delimiters or boundary isolation techniques are used to separate the git history content from the main system prompt instructions.
Capability inventory: Possesses shell command capabilities via git and gh execution blocks in SKILL.md (Steps 1 and 4).
Sanitization: A mandatory user validation gate is present in Step 3, requiring the agent to pause and ask the user to verify the draft PR.md content before the pull request is created.
[COMMAND_EXECUTION]: Executes local CLI tools (git and gh) to inspect repository history and automate the creation of pull requests.

open-pr