trufagent

Use this agent to implement features — React components, TypeScript, Next.js API routes, Prisma queries, CRUD operations, form handling, dashboard UI, Tailwind CSS, external service integrations. Generates complete working code. Use after thinker has designed the approach for complex features. <example> Context: A new UI component is needed. user: "Build the incomplete jobs approval panel component" assistant: "→ delegating to builder for IncompleteApprovalPanel implementation" <commentary> Builder generates complete, production-ready components without placeholders. </commentary> </example> <example> Context: A new API endpoint is needed. user: "Add a PATCH endpoint for updating job status with audit log" assistant: "→ delegating to builder for the API route + Prisma logic" <commentary> Builder handles full API implementation including validation and error handling. </commentary> </example>

Use this agent for deep architectural review on high-stakes changes ONLY — new DB migrations, auth changes, external integrations, irreversible decisions, security-sensitive code. Always AFTER reviewer has already done the first pass. Do NOT use for bug fixes, UI tweaks, or single-file changes. <example> Context: Major integration added after reviewer checked correctness. user: [builder finished Zoho webhook + reviewer found no logic issues] assistant: "→ delegating to critic for architectural review — security, reversibility, coupling" <commentary> Critic focuses on systemic risks that reviewer does not cover. </commentary> </example> <example> Context: Auth flow being modified. user: "We're changing the session token handling in auth.ts" assistant: "→ delegating to critic — auth changes always require architectural review" <commentary> Auth changes are always high-stakes regardless of implementation correctness. </commentary> </example>

Use this agent for all UI/UX design work — new pages, components, mockups, visual redesigns, and design system decisions. Always picks the right skill combination based on what is being designed. Use during brainstorming mockup phase and before builder implements. <example> Context: New marketing or landing page needed. user: "Design a landing page for trufagent" assistant: "→ delegating to designer — landing page: taste-skill + soft-skill" <commentary> Landing pages need creative direction (taste-skill) and premium feel (soft-skill). Designer picks the right combo. </commentary> </example> <example> Context: Dashboard or admin panel design. user: "Design the jobs overview screen for JC-APP" assistant: "→ delegating to designer — dashboard: ui-ux-pro-max + soft-skill" <commentary> Admin panels need UX rules, accessibility, and density guidelines — ui-ux-pro-max covers this. Soft-skill adds polish. </commentary> </example> <example> Context: Existing screen looks generic and needs improvement. user: "This timesheet view looks like a generic AI design, improve it" assistant: "→ delegating to designer — redesign: redesign-skill + ui-ux-pro-max" <commentary> redesign-skill audits the current design and applies premium standards. ui-ux-pro-max validates accessibility. </commentary> </example> <example> Context: Brand-inspired design requested. user: "Make the dashboard feel like Linear" assistant: "→ delegating to designer — brand reference: awesome-design-html + ui-ux-pro-max" <commentary> awesome-design-html has the exact Linear reference with design tokens. ui-ux-pro-max applies the UX rules on top. </commentary> </example>

Use this agent for the first-pass code review after implementation — check logic correctness, error handling, edge cases, TypeScript types, and adherence to project conventions. Use AFTER builder and BEFORE critic. Do NOT use for architectural decisions or security-systemic analysis. <example> Context: builder just implemented a new API route. user: [builder finished POST /api/jobs/[id]/complete] assistant: "→ delegating to reviewer for first-pass review — logic, error handling, types" <commentary> Reviewer catches implementation-level issues that builder may have missed. Claude Sonnet excels at detecting subtle bugs in context. </commentary> </example> <example> Context: A component was updated with new state management. user: [builder updated JobStatusPanel with optimistic updates] assistant: "→ delegating to reviewer for first-pass review — state logic, race conditions, edge cases" <commentary> React state management bugs are subtle — reviewer focuses specifically on correctness. </commentary> </example>

Use this agent for fast, independent tasks where speed matters more than deep reasoning — format conversions, data extraction, running the same operation on multiple inputs, simple lookups. Use when tasks can run in parallel with other agents. <example> Context: User needs multiple files checked simultaneously. user: "Check if all 8 API routes have proper error handling" assistant: "→ delegating to runner to check all routes in parallel" <commentary> Runner excels at repetitive checks across multiple files — no reasoning, just execution. </commentary> </example> <example> Context: Quick data extraction needed. user: "Extract all job IDs with status INCOMPLETE from the logs" assistant: "→ delegating to runner for fast log extraction" <commentary> Low-complexity pattern matching — speed matters, not reasoning depth. </commentary> </example>

Use this skill to reconfigure the model behind a specific agent role in the trufagent fleet. Invoke with /trufagent:config [agent] or when the user says "change [agent] model", "swap [agent] to [model]", "reconfigure [agent]", or "optimize [agent] for cost/speed".

Use this skill to open the trufagent configuration dashboard in the browser. Invoke with /trufagent:dashboard or when the user says "open dashboard", "open trufagent UI", "configure trufagent visually", or "show fleet status".

Use this skill to initialize the trufagent context system in the current project directory — auto-detects the tech stack, asks 5-7 questions about project state, then generates CLAUDE.md and docs/context/ with real content. Run once per project. Invoke with /trufagent:init or when the user says "init trufagent", "initialize project context", or "setup project for trufagent".

Use this skill to set up the trufagent framework globally on this machine — installs LiteLLM, copies agent files, configures hooks, and opens the dashboard for fleet configuration. Run once per machine. Invoke with /trufagent:setup or when the user says "setup trufagent", "configure trufagent", or "install trufagent fleet".

Use this skill to show the current trufagent fleet status — configured agents, models, LiteLLM health, and context system state. Invoke with /trufagent:status or when the user asks "trufagent status", "is litellm running", "what agents are configured", or "show fleet".