security-review-subagent

Run an automated security review on a pull request diff by executing the skill against your branch and base to receive a structured Markdown artifact with file/line citations and suggested fixes. It applies a fixed six-check brief to detect security redlines.

When adding MCP connector grants or agent invocations, check for capability creep, untrusted-input misuse, output exposure, injection-vulnerable shapes, and unscoped capabilities. A fixed six-check security brief flags these specific risk surfaces in the PR diff.

Yes, you can automate prompt injection detection for untrusted input ingestion by running a fresh-context subagent against your PR diff. It flags injection-vulnerable shapes and untrusted-input misuse with structured findings and suggested fixes.

No, security review does not work on documentation-only or CSS-only changes. The skill explicitly declines doc-only and CSS-only PRs, applying its six-check brief only to branches that introduce agent invocations, external surfaces, or untrusted-input ingestion.

A deterministic subagent contract for structured security findings uses a fixed brief template with placeholders for branch and base, keeping results consistent run-to-run. It outputs a Markdown artifact with one section per finding, including citations and a deterministic summary header.

Generate a Markdown security report from a branch diff by running the security review skill. It produces a structured artifact with one section per finding, including file/line citations, risk levels, suggested fixes, and a deterministic summary header.