crew

name	crew
description	Phase-based development workflow manager that guides projects through structured phases from strategy and design through implementation, quality assurance, and launch. Manages roadmaps, feature branches, agent orchestration, and MCP server configuration.

/crew — Phase-Based Development Workflow

Crew manages the full software development lifecycle through structured phases, from onboarding existing codebases to launching production software.

Usage

/crew                         → Show available subcommands
/crew init                    → Initialize Crew in a new project
/crew onboard                 → Onboard an existing codebase (interactive wizard)
/crew setup-mcp               → Configure MCP servers for external tools
/crew setup-mcp jira          → Configure a specific MCP server
/crew feature                 → Add a new feature to the roadmap
/crew feature login           → Quick-add a feature by name
/crew drive                   → End-to-end product-led feature development
/crew drive user-auth         → Quick-start drive for a named feature
/crew gaps                    → Gap-finder audit and fix pipeline
/crew gaps http://localhost   → Audit a specific URL
/crew deploy                  → Pre-deployment checklist with auto-remediation
/crew deploy --phase 1        → Run pre-deploy checks only
/crew deploy --fix            → Auto-dispatch agents to fix failures
/crew features                → Show all features with status and context
/crew features --status done  → Filter features by status/epic/priority
/crew gang-import <slug>      → Import a Gang evaluation as a Crew feature
/crew gang-escalate <slug>    → Send a stalled Crew feature back to Gang for re-evaluation
/crew push                    → Sync the current feature to GitHub Projects (live status card)
/crew resume [id]             → Resume from a usage-budget checkpoint (or list)
/crew profile [check|set|list]→ Manage user knowledge profile per technology
/crew investigate [topic]     → Hypothesis-driven, time-boxed investigation (spike card)
/crew fix [issue]             → Reproduce, locate, fix, verify a known bug (bug card)
/crew performance [target]    → Measure → recommend → optimise → re-measure (with before/after IMPACT.md)

Gang Integration

Crew integrates with the Gang multi-agent committee plugin. Gang evaluates whether to build something; Crew executes the build. See references/gang-integration.md for the full lifecycle.

The two integration commands above (gang-import, gang-escalate) require the Gang plugin installed at ~/.claude/plugins/marketplaces/gang-marketplace/ and a Gang evaluation that has reached at least the advise stage.

Subcommand Router

When the user runs /crew, check the argument to determine which subcommand to execute:

No arguments → Show help

Crew — Phase-Based Development Workflow

Available subcommands:

  /crew init          Initialize Crew in a new project
                          (creates .crew/ state files, sets Phase 1)

  /crew onboard       Onboard an existing codebase to the phase framework
                          (detect project type, import context, audit, set phase)

  /crew setup-mcp     Configure MCP servers for external tools
                          (Jira, Notion, Linear, GitHub, SuperDesign, etc.)

  /crew feature       Add a new feature to the roadmap and start development
                          (from description, external tool link, or existing roadmap)

  /crew drive         End-to-end product-led feature development
                          (pm-architect → implementation → QA → gap audit)

  /crew deploy        Pre-deployment checklist with auto-remediation
                          (verify env → health checks → dispatch agents → update observations)

  /crew gaps          Gap-finder audit and fix pipeline
                          (audit UI → analyze gaps → create fixes → verify)

  /crew features      Feature dashboard — show all roadmap features
                          (status, pages, navigation flow, acceptance progress)

  /crew gang-import   Import a Gang evaluation (GO Package) as a Crew feature
                          (requires the Gang plugin and a completed evaluation)

  /crew gang-escalate Escalate a stalled or invalidated Crew feature back to Gang
                          (for re-scoring when scope or assumptions change)

  /crew push          Sync the current feature to GitHub Projects v2
                          (live status card, updates Status field per phase)

  /crew resume        Resume from a usage-budget checkpoint
                          (auto-fired by scheduled tasks; can run manually)

  /crew profile       Manage user knowledge profile per technology
                          (one-time per tech; cached; sets plain-English mode
                           when any tech is none/low)

  /crew investigate   Hypothesis-driven, time-boxed investigation
                          (creates spike card; read-only agents; ends with
                           a recommendation. Technical only — for strategic
                           investigations use /gang directly)

  /crew fix           Reproduce, locate, fix, verify a known bug
                          (creates bug card; reproduce → code-reviewer locate
                           → engineer dispatch → qa-engineer verify)

  /crew performance   Measure → recommend → optimise → re-measure
                          (Chrome DevTools MCP for runtime metrics,
                           static analysis fallback; before/after
                           IMPACT.md with numeric headline metrics)

Current project status:
  → Read .crew/current-phase.yaml if it exists, show phase + project type
  → If missing: "No project onboarded. Run /crew onboard to get started."

init → Load project initializer

Read and follow skills/crew/commands/init.md

onboard → Load onboarding wizard

Read and follow skills/crew/commands/onboard.md

setup-mcp → Load MCP setup helper

Read and follow skills/crew/commands/setup-mcp.md Pass any additional arguments (e.g., jira, notion) as the tool name.

feature → Load feature command

Read and follow skills/crew/commands/feature.md Pass any additional arguments (e.g., login) as the feature name for quick-add.

drive → Load drive command

Read and follow skills/crew/commands/drive.md Pass any additional arguments (e.g., user-authentication) as the feature name for quick-start.

deploy → Load deploy command

Read and follow skills/crew/commands/deploy.md Pass any additional arguments (e.g., --phase 1, --fix, --skip-build, --skip-tests) as flags.

gaps → Load gaps command

Read and follow skills/crew/commands/gaps.md Pass any additional arguments (e.g., http://localhost:5173) as the target URL. Also pass any flags (--quick, --a11y, --responsive, --compare {url}).

features → Load features command

Read and follow skills/crew/commands/features.md Pass any flags (--status, --epic, --priority) as filters.

gang-import → Load Gang import command

Read and follow skills/crew/commands/gang-import.md Pass the Gang evaluation slug as the argument. Dispatches the gang-bridge agent.

gang-escalate → Load Gang escalation command

Read and follow skills/crew/commands/gang-escalate.md Pass the Crew feature ID as the argument. Dispatches the gang-bridge agent.

push → Load GitHub Projects sync command

Read and follow skills/crew/commands/push.md Creates a draft card on GitHub Projects v2 board(s) for the current feature, or updates the existing card's Status field based on the current phase. Pass --dry-run to preview without making API calls. Pass a feature ID as positional arg to push a specific feature instead of the current one.

resume → Load resume-from-checkpoint command

Read and follow skills/crew/commands/resume.md Resumes a feature build that was paused by the usage-budget system. With no argument, resumes from the most recent checkpoint. Pass a checkpoint ID for a specific one, --list to see all checkpoints, or --discard {id} to mark one as won't-resume. This is idempotent — concurrent invocations on the same checkpoint are safe.

profile → Load user-profile command

Read and follow skills/crew/commands/profile.md Manages the user's per-technology knowledge profile. Subcommands: check (default — detects techs and asks about new ones), set <tech> <level>, list, clear <tech>, reset. When any cached tech is none/low, all subsequent agent dispatches get a "PLAIN ENGLISH MODE" prefix.

investigate → Load investigation command

Read and follow skills/crew/commands/investigate.md Hypothesis-driven investigation, time-boxed. Creates a spike card on the GitHub Projects board. Read-only agent dispatch — no code changes during investigate. Always ends with a recommended next action, calibrated to confidence: HIGH suggests a specific next command (/crew fix or /crew feature) with context pre-filled; MEDIUM suggests deepening the investigation or adding observability; LOW suggests re-running with a different --type or escalating to /gang. The recommendation is the first option (marked ★) but never the only one — user can always pick differently. Technical investigations only — /gang handles strategic ones.

performance → Load performance command

Read and follow skills/crew/commands/performance.md End-to-end performance loop: measure baseline → analyse → recommend prioritised fixes → optionally implement → re-measure → write a numeric impact report. Detects Chrome DevTools MCP at runtime — uses Lighthouse + perf trace + network waterfall + memory snapshots when available; static-analysis fallback when not. Budget profiles (strict / standard / relaxed) drive Web Vitals thresholds and pass/fail oracles. Defaults to "ask after measurement" before implementing; flags --no-implement and --auto-implement-quick-wins override. IMPACT.md is numeric-first: before/after metrics tables, deferred items auto-spawn follow-up spike, regression beyond 10% tolerance halts the run.

fix → Load bug-fix command

Read and follow skills/crew/commands/fix.md Reproduce → locate → fix → verify a known bug. Creates a bug card. Skips Crew Phase 1 (Strategy) and Phase 2 (Design). Locate uses code-reviewer in read-only mode AND scans the codebase for the same pattern in other places (EXACT / FUZZY / LOOSE classification). Step 4.5 confirms scope: fix all matches / review each / primary only / EXACT only — sites NOT included get auto-spiked for later review. Fix uses ui-engineer / api-engineer; cross-layer bugs ask sequential vs parallel dispatch. Verify uses qa-engineer; the fix MUST satisfy the regression test the command writes during the Reproduce step. After verify, Step 7.5 generates .crew/fixes/{bug_id}/IMPACT.md — a structured report covering files modified, public interfaces changed, blast radius, behavior changes, risk areas (deferred sites + adjacent code), production monitoring suggestions, rollback plan, test coverage delta, and a sign-off checklist for the human reviewer.

Unknown subcommand

Unknown subcommand: [arg]

Available: init, onboard, setup-mcp, feature, drive, deploy, gaps, features, gang-import, gang-escalate, push, resume, profile, investigate, fix, performance
Run /crew for help.