The Jagoda Toolkit

Review changed files for tech debt lint can't catch

Run lint, typecheck, test on affected workspaces only. Auto-detects package manager.

Write tests for code changes while preserving harmonic constraints, matching local test patterns, and verifying with format/lint/typecheck/test. <example> Context: User wants tests for code they just wrote. user: "Write tests for the classifier service" assistant: "I'll use the test-writer agent to create tests for the classifier service." <commentary> User explicitly asking for tests for a specific file. </commentary> </example> <example> Context: New module was added and needs test coverage. user: "Add tests for the user schemas" assistant: "I'll use the test-writer agent to write tests matching existing patterns." <commentary> Test coverage needed for a new module. Trigger test-writer. </commentary> </example>

Checks $PAGE UI in Chrome via DevTools MCP. Reports errors — never fixes anything. <example> Context: User wants to check the page UI works in Chrome. user: "Check the UI for the dashboard" assistant: "I'll use the ui-verifier agent to navigate to the page, test submissions with real data, and report any issues." <commentary> UI-only verification. User wants Chrome check, not CI or DB. </commentary> </example> <example> Context: User checked the CI, now wants UI verification. user: "CI passed, now check the UI for the settings page" assistant: "I'll use the ui-verifier agent to verify the settings page in Chrome." <commentary> User is running agents individually in sequence. UI check after CI pass. </commentary> </example>

Deep architecture audit for any feature or change. Runs failure mode analysis, boundary audit, test gap mapping, contract compliance, data flow tracing, and risk prioritization. Use for major features, critical refactors, or when things feel architecturally wrong.

Structured task handoff. Use when user wants to plan or set up a task before executing — asks 5 intake questions (mode, goal, context, format, unknown_ok) and confirms in one sentence before starting.

Launch a named expert protocol. Usage: /protocol <name> <task>. Run /protocol with no args to list available protocols.

AI-agent compatibility audit. Use when evaluating whether code changes are safely readable, analyzable, and modifiable by AI agents. Checks determinism, explicitness, complexity, and boundary clarity.

Production-readiness audit for authentication. Use when auth code changes or when auth feels fragile, unclear, or unsafe. Covers OIDC, sessions, tokens, route protection, and secret management.

6 hooks across 3 events