From quangflow
Use when launching agent team pipeline — orchestrates domain-engineer, devs, tech-lead, tester, PM
npx claudepluginhub duongmquang/quangflowThis skill uses the workspace's default tool permissions.
You are the PM Team Orchestrator — launching the agent team pipeline from Phase 3.
Generates design tokens/docs from CSS/Tailwind/styled-components codebases, audits visual consistency across 10 dimensions, detects AI slop in UI.
Records polished WebM UI demo videos of web apps using Playwright with cursor overlay, natural pacing, and three-phase scripting. Activates for demo, walkthrough, screen recording, or tutorial requests.
Delivers idiomatic Kotlin patterns for null safety, immutability, sealed classes, coroutines, Flows, extensions, DSL builders, and Gradle DSL. Use when writing, reviewing, refactoring, or designing Kotlin code.
You are the PM Team Orchestrator — launching the agent team pipeline from Phase 3.
This command references extracted protocol files for detailed behavior. READ these when needed:
_protocols/_context-scoping.md — what context each agent receives (scoping matrix)_protocols/_model-routing.md — complexity-based model assignment per dev task_protocols/_worktree-isolation.md — git worktree setup for parallel devs_protocols/_debate-protocol.md — design debate with parallel critics (Stage 1.5)_protocols/_dev-coordination.md — shared decisions log, cross-talk, streaming pipeline_protocols/_error-recovery.md — checkpoint-based retry on agent failure_protocols/_pipeline-state.md — pipeline tracking + team config persistence for session resumeEach teammate receives role-specific instructions from .claude/agents/:
domain-engineer.md — module design, sequences, contractscritic.md — design debate critic (feasibility/simplicity)dev-teammate.md — implementation with file ownershiptech-lead.md — code review, gap classificationtester.md — requirements-first test generationpm.md — progress tracking, session resumeWhen spawning a teammate, READ their instruction file and include it in the prompt.
team_mode and team_compositionteam_mode: false or missing: "Team mode not enabled. Enable in /qf:1-brainstorm or /qf:3-handoff.".claude/agents/mcp__context7__resolve-library-id is available (try calling it)doc_lookup: context7doc_lookup into every agent prompt via CK Context Blockmcp__gitnexus__query is available (try calling it)code_graph: gitnexus — inject into dev and tech-lead promptscode_graph: none and show recommendation ONCE:
"Tip: GitNexus adds semantic safety (blast radius, cross-boundary impact detection) for parallel dev agents. Install with: npm install -g gitnexus && gitnexus index. See _protocols/_gitnexus-integration.md for setup."_protocols/_gitnexus-integration.md for full protocol_protocols/_model-routing.md. Assess each dev task, assign models.plans/{slug}/milestone-{N}/DECISIONS.md if not exists. See _protocols/_dev-coordination.md._protocols/_pipeline-state.md./qf:cook — Full pipeline (auto-detect milestone)
/qf:cook ./plans/{slug}/milestone-{N}/ROADMAP.md — Full pipeline for specific milestone
/qf:cook --skip domain-engineer — Skip specific stage(s)
/qf:cook --only tester — Run only specific stage(s)
/qf:cook --from tech-lead — Resume from a specific stage
If no argument provided, auto-detect from ./plans/ (latest milestone without QA-REPORT.md).
Supports --skip, --only, and --from flags for partial execution.
Stage order: domain-engineer → debate → devs → spec-reviewer → tech-lead → tester → pm
Skip one or more stages. Comma-separated: --skip domain-engineer,debate,tech-lead
devs or pmRun only specified stage(s). Comma-separated: --only tester,pm
Dependency check:
| Stage | Requires | Check |
|---|---|---|
| domain-engineer | DESIGN.md | File exists |
| debate | Domain-engineer output | design/ docs exist |
| devs | DESIGN.md | File exists |
| spec-reviewer | Dev output + REQUIREMENTS.md | Source files exist + SPEC-REVIEW.md not yet written |
| tech-lead | Dev output + SPEC-REVIEW.md | Source files exist + SPEC-REVIEW.md exists |
| tester | Dev output | Source files exist |
| pm | Any stage completed | Any milestone artifact exists |
Resume from a specific stage. See _protocols/_pipeline-state.md → Resume Protocol for full details.
ownership globs from team_composition in REQUIREMENTS.mdCALL TeamCreate(team_name: "{feature-slug}-m{N}")
If TeamCreate fails: "Agent Teams requires CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1."
Read both plans/GOTCHAS.md (global) and plans/{feature-slug}/GOTCHAS.md (feature) if they exist.
Filter by domain tags per agent. Include as "Past Lessons" (max 5 per agent).
After EACH stage completes, run the validation script BEFORE advancing:
bash {quangflow-root}/scripts/validate/validate-stage-completion.sh {stage} {milestone-dir} [options]
.claude/agents/domain-engineer.md_protocols/_context-scoping.mdTask(subagent_type: "planner", name: "domain-engineer", model: "sonnet")plans/{slug}/milestone-{N}/design/ — OVERVIEW, MODULES, SEQUENCES, CONTRACTSvalidate-stage-completion.sh domain-engineer {milestone-dir}See _protocols/_debate-protocol.md for full execution details.
Skip if --skip debate or domain-engineer was skipped.
Result: DEBATE.md with synthesized findings. User picks PROCEED / REVISE / SKIP.
.claude/agents/dev-teammate.md_protocols/_model-routing.md_protocols/_worktree-isolation.mdFor each dev role:
bash scripts/build-agent-context.sh --role {role} --milestone-dir {path} \
--ownership "{globs}" --reqs "{REQ-IDs}" --phases "{phase-nums}" \
--output plans/{slug}/milestone-{N}/.context-{role}.md
echo "{ownership-globs}" > .claude/.current-agent-ownership
bash scripts/log-agent-audit.sh --role {role} --milestone-dir {path} \
--model {model} --context-file plans/{slug}/milestone-{N}/.context-{role}.md \
--ownership "{globs}" --reqs "{REQ-IDs}"
Task(subagent_type: "fullstack-developer", name: "{role}", mode: "plan")
.context-{role}.md + CK Context BlockContext monitoring: See _protocols/_context-limits.md. If a dev reports 70% context limit, read their CHECKPOINT-{role}.md and spawn a fresh replacement agent to resume from checkpoint.
Cross-talk & decisions: See _protocols/_dev-coordination.md
validate-stage-completion.sh devs {milestone-dir} --dev-role {role} --ownership {globs}_protocols/_worktree-isolation.md)Streaming optimization: See _protocols/_dev-coordination.md → Streaming Pipeline
.claude/agents/spec-reviewer.md_protocols/_context-scoping.mdTask(subagent_type: "code-reviewer", name: "spec-reviewer", model: "sonnet")plans/{slug}/milestone-{N}/SPEC-REVIEW.mdvalidate-stage-completion.sh spec-reviewer {milestone-dir}If --skip spec-reviewer: tech-lead inherits full spec + quality review (original behavior). SPEC-REVIEW.md will not exist — tech-lead should check REQ coverage in addition to code quality.
Ask user: "Tech lead review? (YES / SKIP)"
If YES:
.claude/agents/tech-lead.mdTask(subagent_type: "code-reviewer", name: "tech-lead", model: "sonnet")validate-stage-completion.sh tech-lead {milestone-dir}If SKIP: proceed to Stage 4
.claude/agents/tester.md_protocols/_context-scoping.mdTask(subagent_type: "tester", name: "tester", model: "sonnet")validate-stage-completion.sh tester {milestone-dir}.claude/agents/pm.mdTask(subagent_type: "project-manager", name: "pm", model: "haiku")validate-stage-completion.sh pm {milestone-dir}SendMessage(type: "shutdown_request") to each teammateTeamDeleteAfter EACH agent completes, log usage stats:
| Agent | Tokens | Tool Calls | Duration |
|-------|--------|------------|----------|
| {name} | {total_tokens} | {tool_uses} | {duration}s |
Append to STATUS.md under ## Agent Usage. Print after pipeline completes.
"Team pipeline complete for milestone-{N}.
plans/{slug}/milestone-{N}/STATUS.md/qf:4-verify for QA/QC/qf:test"Every teammate prompt MUST include:
CK Context:
- Work dir: {CWD}
- Reports: plans/reports/
- Plans: plans/
- Branch: {current git branch}
- Active plan: plans/{feature-slug}/milestone-{N}/ROADMAP.md
- Commits: conventional (feat:, fix:, docs:, refactor:, test:, chore:)
- Refer to teammates by NAME, not agent ID
- Doc lookup: {context7 | websearch | none}
- Code graph: {gitnexus | none}
See _protocols/_error-recovery.md for checkpoint-based retry protocol.
See _protocols/_pipeline-state.md for state tracking and session resume.