The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard git commands (git diff, git status) to perform its primary function of inspecting code changes. These are executed within the user's repository context for the intended functionality of code review.- [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it processes untrusted data from code diffs and pull request descriptions. Evidence chain: (1) Ingestion points: The agent ingests data from git diff output and user-provided PR/issue descriptions as defined in the 'Understand Intent' and 'Review Implementation' sections of SKILL.md. (2) Boundary markers: The skill includes a structured 'Reviewer Prompt' template that explicitly instructs the agent to 'Read the actual diff/code' and 'Do not trust the author's summary'. (3) Capability inventory: Subprocess calls to git are specified in SKILL.md. (4) Sanitization: No explicit text sanitization or escaping is performed on the ingested diff content. This represents a standard risk for code review tasks and is addressed by the instructional boundaries provided in the skill.

doing-code-review