Deterministic design-system linter for rendered websites. CLI + MCP server for AI coding agents.
Agent Claude Code
Implements approved specs in the Plumb Rust workspace. TDD, atomic commits, workspace layer rules, strict lint policy. Use for any non-trivial code change; skip for trivial typo/comment/label edits.
Agent Claude Code
First-gate reviewer. Checks implementation against the spec or ticket only — not code style. Use after the implementer agent finishes.
Agent Claude Code
Second-gate reviewer. Checks Rust idioms, error handling, lint compliance, and workspace discipline. Use after 02-spec-reviewer approves.
Agent Claude Code
Final gate. Runs the full test suite and reports pass/fail with structured summary. Use after code-quality review approves.
Agent Claude Code
Enforces the workspace dependency hierarchy and invariants. Use when a change touches crate boundaries, Cargo.toml dependencies, or adds a new crate.
Agent Claude Code
Security-focused review pass. Use on PRs that touch input parsing, the MCP server surface, URL handling, or dependency graph changes.
Agent Claude Code
Root-cause analysis for failing tests, panics, or CI regressions. Use when a build is broken and you need a diagnosis before the implementer touches code.
Agent Claude Code
Authors a new Plumb rule from a specification. Use when adding a rule in crates/plumb-core/src/rules/. Follows TDD and the cookie-cutter pattern.
Agent Claude Code
Authors a new MCP tool in crates/plumb-mcp. Follows the compact-response contract and protocol test pattern. Use when extending the MCP surface.
Agent Claude Code
Rapid minimal fix for trivial issues (typos, lint drift, snapshot churn, dep bumps). Use for changes that don't need TDD or a spec — otherwise use 01-implementer.