Agent Claude Code
Given a list of security findings, identifies chains where one finding enables another and writes a plain-English attack narrative. Invoked after vulnerability-audit and design-review have returned.
Agent Claude Code
Audits one code hotspot for a contract mismatch — a divergence between what callers assume the hotspot guarantees and what the hotspot body actually enforces. Returns hypotheses tagged VERIFIED, REFUTED, or NEEDSMORE. Invoked once per hotspot per round by the contract-review orchestrator.
Agent Claude Code
Audits a codebase for missing security controls — the gaps that pattern-matching auditors won't catch, like no timeout, no cost cap, no rate limit, prose-only guards. Invoked in parallel with vulnerability-audit calls.
Agent Claude Code
Second-pass refutation filter for security-review findings. Reads each candidate finding's cited code and drops the ones with concrete refutation evidence (a guard, middleware, sanitizer, or correct API call at the cited location). Bias is toward keeping; uncertain findings pass through.
Agent Claude Code
Finds security-sensitive code locations in a repository — the files and functions a reviewer should look at. Reads the threat model for context, then enumerates and ranks hotspots. Invoked after threat-modeling, before per-hotspot review.
Agent Claude Code
Builds a threat model for a codebase — what it does, where it runs, which inputs are trusted, which are untrusted. Pure context for downstream subagents; does not decide which files to look at. Invoked at the start of a security review or when planning a new feature that handles user data.
Agent Claude Code
Audits all hotspots in a single file for vulnerabilities by applying named Soundcheck skills. Returns findings in plain language a non-security developer can act on. Invoked once per file (with that file's full hotspot list).