The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local repository which could theoretically contain malicious instructions designed to influence the agent.
Ingestion points: Untrusted data enters the agent context via git status, git diff, and git log commands executed in SKILL.md (Workflow Step 1).
Boundary markers: The skill implements strong boundary markers in SKILL.md (Workflow Step 6), which requires the agent to print the file list and the generated message for user review before proceeding with the commit.
Capability inventory: The skill has capabilities to read repository metadata and write a single commit to the current branch via git CLI operations.
Sanitization: The skill includes an explicit 'Security Model' section in SKILL.md that instructs the agent to treat all tool output, files, and logs as data rather than instructions, specifically forbidding these inputs from authorizing additional actions like pushing or amending.
[SAFE]: The skill uses git commit -F - with a heredoc to prevent shell injection or escaping issues that can occur when passing messages via the -m flag.
[SAFE]: The workflow includes a specific step to screen untracked files for potential secrets or sensitive scripts before they are staged, reducing the risk of accidental data exposure.

commit-all