The Agent Skills Directory

[PROMPT_INJECTION]: Potential for Indirect Prompt Injection. The skill summarizes data from git logs and code diffs which are untrusted external inputs.
Ingestion points: SKILL.md (Step 2: outputs of git log, git diff --stat, and git diff).
Boundary markers: Absent; the skill prompt does not define clear delimiters for the ingested git data during the drafting phase.
Capability inventory: Capabilities include file system writing (Write), branch pushing (git push), and pull request creation (gh pr create).
Sanitization: The skill mitigates risks by requiring user confirmation via AskUserQuestion before final execution and by using the Write tool to store the PR body in a temporary file to prevent direct shell injection.
[COMMAND_EXECUTION]: Executes local system commands via the Bash tool. The skill relies on standard git and gh (GitHub CLI) tools to inspect repository state and automate the PR workflow.

send-it