The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from code diffs and file contents. This is a standard functional requirement for a code review tool, but it introduces a surface for indirect prompt injection if the content being reviewed contains instructions intended to influence the agent.
Ingestion points: Git diff outputs and source file contents retrieved in the 'Obtain changes' step (SKILL.md).
Boundary markers: The instructions do not define specific delimiters to separate code from instructions or advise the agent to ignore embedded commands within the diffs.
Capability inventory: The skill executes git commands for diffing and logging (git-commands.md).
Sanitization: No sanitization or filtering of the ingested code content is specified before analysis.
[COMMAND_EXECUTION]: The skill relies on executing git commands to perform its core functions. These commands are used to deduce intent, triage the scope of the review, and fetch code changes. The use of these commands is consistent with the skill's stated purpose and is limited to read-only repository operations.

review-code-changes