The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/code-review.cjs executes local system binaries, specifically git for diffing and xmllint for XML validation.
[COMMAND_EXECUTION]: The skill dispatches prompts to various local AI CLI tools (including claude, gemini, copilot, and cursor-agent) and instructs them to perform tasks. This includes a 'readiness probe' where the skill directs another agent to execute a shell command to verify connectivity.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository, such as git diffs and plan documents. These files could contain instructions designed to manipulate the behavior of the reviewer agent.
Ingestion points: Cumulative diffs and plan documents are read from the workspace using git diff and standard file reads in scripts/code-review.cjs.
Boundary markers: The skill uses explicit delimiters like <<<BEGIN CUMULATIVE DIFF>>> and <<<BEGIN REVIEW XML>>> to isolate untrusted content and model outputs.
Capability inventory: The skill can execute various AI CLI tools and system commands via child_process.spawn and execFileSync.
Sanitization: The skill validates the XML output using xmllint with the --nonet flag to prevent external entity resolution and ensure compliance with a local schema.

st-code-review