The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local git and gh (GitHub CLI) commands to manage the development lifecycle. It also runs a bundled bash script, estimate-base-branches.sh, which uses standard git utilities to calculate branch merge bases. No unauthorized or suspicious command patterns were detected.
[DATA_EXFILTRATION]: The skill uploads data to the remote origin via git push. To prevent accidental exposure of secrets, the workflow includes a safety check that monitors for sensitive file names (e.g., .env, credentials, .pem) and halts the automation for mandatory user confirmation if such files are part of the changes.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection (Category 8) because it uses repository content (diffs and logs) to generate the PR draft.
Ingestion points: External data enters through git diff and git log output processed in Step 3.
Boundary markers: None; the raw git output is interpolated directly into the context for summary generation.
Capability inventory: The skill has permissions to write to the file system (git commit) and perform network operations (git push/gh pr create).
Sanitization: There is no automated sanitization of the diff content, but the skill enforces a mandatory preview and approval step in Step 4, ensuring a human verifies the generated content before it is published.

create-pr