The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands such as git log, git diff, and git status to retrieve commit history and code changes from the local repository. It also optionally leverages the sem CLI for semantic diffing and impact analysis. These operations are restricted to read-only access and are standard for development workflows.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository, including commit messages and file diffs. While this provides a surface for indirect prompt injection, the risk is assessed as safe due to the absence of high-privilege capabilities such as network exfiltration or file system modifications. Evidence chain: 1. Ingestion points: Git history and file diffs via git and sem commands. 2. Boundary markers: None explicitly used to delimit untrusted git data. 3. Capability inventory: Read-only git operations and Semantic CLI; no network or write access. 4. Sanitization: No specific sanitization or filtering of commit messages or code contents.

pr-description