The Agent Skills Directory

[SAFE]: The skill uses local repository tools (git, rg) for analysis as part of its primary function. No evidence of data exfiltration, credential harvesting, or unauthorized remote access was found.
[COMMAND_EXECUTION]: The skill uses a limited set of shell commands (git status, git diff, git log, rg) to inspect the current repository. These are appropriate for the tool's stated purpose of auditing code drift and domain modeling.
[DATA_EXPOSURE]: The skill performs repository-wide searches (rg) and log inspections. While this exposes codebase content to the AI agent, it is an intended behavior for the review process and does not involve sending data to external endpoints.
[INDIRECT_PROMPT_INJECTION]: Because the skill processes external data (code, diffs, and commit messages), it is susceptible to indirect prompt injection if malicious instructions are present in the repository being analyzed. However, the skill lacks high-risk capabilities such as network exfiltration or arbitrary command execution that would allow an attacker to exploit this surface effectively.
Ingestion points: git diff, git log, rg, and file read operations.
Boundary markers: Absent; the skill does not wrap untrusted repository data in specific delimiters.
Capability inventory: The agent is limited to standard repository inspection commands; no eval, curl, or file-write operations are requested.
Sanitization: No explicit sanitization or filtering of repository content is performed before processing.

theory-building-review