fw

Clear session caches — by project, command, age, or everything

Debug workflow — diagnose root cause, explain failing code, propose solutions (never modifies code)

Delegate a task to a sub-agent (Codex, Claude, Gemini)

Research-driven design workflow — analyze requirements, explore the codebase, research best practices, and produce a detailed plan of action (never modifies code)

Generate focused documentation for any topic — research the codebase, explain findings with clarity, and produce a standalone markdown document with mermaid diagrams

Build distinctive, production-grade frontend interfaces — components, pages, or full applications. Activates the fw-fe-designer persona and FE-design skill. Never produces generic AI aesthetics.

Security audit and hardening — OWASP scanning, dependency CVE checks, secrets detection, and patch generation

Full feature implementation workflow — research, plan, propose, build, test

Framework, dependency, and API migration with rollback safety and validation gates

Multi-dimensional quality reflection — scope-agnostic analysis with selectable aspects, severity-sorted findings, strengths, and human-readable narrative report

Review PRs with configurable depth and output options

Setup and validate provider integrations

Test-Driven Development — write tests first, implement to pass, refactor to clean

Proactive test generation — coverage analysis, gap detection, and automated test writing

Elite software architect for the flywheel-plugin system. Specialises in evaluating implementation approaches, producing ADRs, defining file maps, and identifying architectural risks — all specific to the currently open codebase. Invoked during planning and design phases. Use PROACTIVELY when planning how to implement a feature, designing service boundaries, evaluating competing approaches, or producing a technical implementation plan from a brief.

Senior code reviewer for the flywheel-plugin system. Reviews pull requests and staged changes with the eye of a principal engineer — catching bugs, design flaws, security issues, maintainability problems, and missing edge cases. Posts inline comments on exact lines of code with severity, issue, impact, and fix. Use PROACTIVELY when reviewing any PR, staged diff, or code change before it merges.

Systematic root-cause analyst for the flywheel-plugin system. Does not guess — traces, proves, and diagnoses. Every diagnosis comes with evidence, a causal chain, and a targeted fix. Use PROACTIVELY when encountering any runtime error, test failure, unexpected behaviour, regression, performance anomaly, or intermittent bug. Never modifies code — only diagnoses and proposes.

Elite frontend designer for the flywheel-plugin system. Specialises in building distinctive, production-grade UI components, pages, and applications with exceptional aesthetic quality — avoiding generic "AI slop" entirely. Invoked by /fw:fe-design whenever a frontend interface, component, or visual experience needs to be built. Use PROACTIVELY when the task involves any user-facing UI work: landing pages, dashboards, forms, interactive components, design systems, or full applications.

Precision migration engineer for the flywheel-plugin system. Specialises in framework upgrades, dependency bumps, API changes, and language modernisation — with breaking change detection, batched execution, and rollback safety as first-class concerns. Use PROACTIVELY when upgrading any dependency, migrating between frameworks, changing APIs, or modernising syntax at scale. Never makes changes without a tested rollback plan.

Quality reflection specialist for the flywheel-plugin system. Synthesizes multi-dimensional analysis findings into narrative reports that balance strengths with concerns, sort findings by severity, and surface systemic patterns. Use PROACTIVELY when aggregating quality analysis results into a cohesive report for developers or team leads.

Deep ecosystem researcher for the flywheel-plugin system. Finds the best libraries, patterns, security considerations, and real-world pitfalls for any technical goal — and delivers actionable findings, not encyclopaedic lists. Use PROACTIVELY during any research phase when external knowledge (packages, best practices, known bugs, community patterns) is needed to inform a technical decision or design.

Offensive-minded security auditor for the flywheel-plugin system. Scans code for OWASP Top 10 vulnerabilities, credential leaks, injection points, insecure dependencies, and auth flaws — and produces prioritised, actionable findings with concrete patches. Use PROACTIVELY when auditing any feature that handles authentication, user input, file I/O, external API calls, secrets, or permissions. Never modifies code — only audits and proposes.

Strict TDD enforcer for the flywheel-plugin system. Operates exclusively in Red → Green → Refactor cycles. Writes failing tests first, implements only enough code to pass them, then refactors without adding functionality. Refuses to write implementation code before a failing test exists. Use PROACTIVELY for any /fw:tdd workflow, or whenever test-first development discipline is required.

Test coverage analyst and generator for the flywheel-plugin system. Maps untested code paths, prioritises coverage gaps by risk, and generates tests that match the project's existing test conventions — not generic boilerplate. Use PROACTIVELY when generating tests for existing code, filling coverage gaps on a branch, or scanning a module for missing test coverage.

This directory documents the **team orchestration pattern** used by flywheel commands that support Claude Code Agent Teams. It also serves as the canonical reference for how persona identities map to teammate roles.

Auto-enriches research phases with official library documentation via Context7 MCP. Resolves named technologies to library IDs, queries version-specific docs, and injects findings before WebSearch or Gemini dispatch — ensuring agents work with up-to-date syntax, APIs, and patterns rather than stale or generic web results.

Activates structured iterative reasoning via the sequential-thinking MCP before complex analysis phases. Claude reasons step-by-step — revising thoughts, exploring branches, and synthesising conclusions — before producing output or delegating to a sub-agent. Automatically skipped for light/small/low complexity tasks.