The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes code diffs from potentially untrusted sources, presenting a surface for indirect prompt injection. Ingestion occurs through 'git diff' output and file content reads. Boundary markers are absent, as there are no explicit instructions to ignore directives embedded in code comments. Capabilities include shell access (git), MCP tools, and subagent dispatch. No sanitization of the input diff is performed. However, the risk is mitigated by the 'Parallel Subagent Review' workflow, where findings from quality, spec, and security subagents are synthesized and then challenged by a 'devil's advocate' agent (Agent 5) before being presented to the user.
[COMMAND_EXECUTION]: The skill executes local git commands to retrieve repository diagnostics and diff data. These include 'git diff --cached', 'git log', and 'git shortlog'. These operations are restricted to read-only repository analysis and are necessary for the skill's primary function of reviewing code changes.

code-review