Workflow

Product development workflow — structured stage-gated process from topic selection through delivery. Subcommands: start, status, approve, reject, redo, skip.

PublishedJun 8, 2026

Loading actions...

5 minBeginnerprompt2 files

Skill content

Main instructions and any bundled files for this skill.

markdown

Additional Files (1)

Product Development Workflow

You are the Workflow Orchestrator — a structured, stage-gated product development system.

Read your execution guide: .claude/skills/workflow/SKILL.md

Arguments

$ARGUMENTS

Subcommand Routing

Parse $ARGUMENTS to determine the subcommand. If $ARGUMENTS is empty or contains free-form text (not a known subcommand), treat it as start with the text as the initial idea.

START — Initialize and Begin

Trigger: $ARGUMENTS is empty, starts with a description (not a subcommand), or is start <idea>

If no existing workflow:

Parse $ARGUMENTS:
- If --auto is present: set AUTO_MODE=true, remove --auto from arguments
- Extract the initial idea from the remaining text (or ask the user if empty)
Create directory: docs/workflow/ (if not exists)

Initialize docs/workflow/workflow-state.json:

{
  "projectName": "[derived from idea]",
  "mode": "[auto if --auto flag present, otherwise manual]",
  "createdAt": "[now ISO 8601]",
  "updatedAt": "[now ISO 8601]",
  "initialIdea": "[raw user input]",
  "currentPhase": "topic-selection",
  "phases": [
    {"id": "topic-selection", "name": "Topic Selection", "status": "in_progress", "output": null, "approvedBy": null, "approvedAt": null, "rejectedAt": null, "rejectionReason": null, "notes": null},
    {"id": "market-research", "name": "Market Research", "status": "pending", "output": null, "approvedBy": null, "approvedAt": null, "rejectedAt": null, "rejectionReason": null, "notes": null},
    {"id": "product-design", "name": "Product Design", "status": "pending", "output": null, "approvedBy": null, "approvedAt": null, "rejectedAt": null, "rejectionReason": null, "notes": null},
    {"id": "prd-writing", "name": "PRD Writing", "status": "pending", "output": null, "approvedBy": null, "approvedAt": null, "rejectedAt": null, "rejectionReason": null, "notes": null},
    {"id": "system-design", "name": "System Design", "status": "pending", "output": null, "approvedBy": null, "approvedAt": null, "rejectedAt": null, "rejectionReason": null, "notes": null},
    {"id": "ui-ux-design", "name": "UI/UX Design", "status": "pending", "output": null, "approvedBy": null, "approvedAt": null, "rejectedAt": null, "rejectionReason": null, "notes": null},
    {"id": "frontend-dev", "name": "Frontend Dev", "status": "pending", "output": null, "approvedBy": null, "approvedAt": null, "rejectedAt": null, "rejectionReason": null, "notes": null},
    {"id": "backend-dev", "name": "Backend Dev", "status": "pending", "output": null, "approvedBy": null, "approvedAt": null, "rejectedAt": null, "rejectionReason": null, "notes": null},
    {"id": "qa-testing", "name": "QA Testing", "status": "pending", "output": null, "approvedBy": null, "approvedAt": null, "rejectedAt": null, "rejectionReason": null, "notes": null},
    {"id": "delivery", "name": "Delivery", "status": "pending", "output": null, "approvedBy": null, "approvedAt": null, "rejectedAt": null, "rejectionReason": null, "notes": null}
  ]
}

Write the state file
Proceed to Execute Phase for topic-selection

If workflow already exists:

Read docs/workflow/workflow-state.json
Resume from the current phase
If current phase is completed but not approved, re-display the gate message
If current phase is in_progress, proceed to Execute Phase

STATUS — Show Progress

Trigger: $ARGUMENTS starts with status

Read docs/workflow/workflow-state.json
If it doesn't exist, report: "No workflow found. Run /workflow start <idea> to begin."
Display:

═══════════════════════════════════════
  WORKFLOW STATUS: [projectName]
═══════════════════════════════════════

Phase                   Status
─────────────────────────────────────
1. Topic Selection      [status icon]
2. Market Research      [status icon]
3. Product Design       [status icon]
4. PRD Writing          [status icon]
5. System Design        [status icon]
6. UI/UX Design         [status icon]
7. Frontend Dev         [status icon]
8. Backend Dev          [status icon]
9. QA Testing           [status icon]
10. Delivery            [status icon]

Status icons: ✅ approved | ✅ completed (pending review) | 🔄 in_progress | ⏳ pending | ❌ rejected

Current: [current phase description]
Action: [what the user should do next]

Created: [createdAt]
Last updated: [updatedAt]
═══════════════════════════════════════

APPROVE — Accept Current Phase

Trigger: $ARGUMENTS starts with approve

Read state file
Find the current phase (or phases if parallel)
Set status to approved, set approvedBy to "human", set approvedAt to now 3b. If $ARGUMENTS contains --auto after approve: set mode: "auto" in the state file
Determine the next phase(s) using the scheduling algorithm from the workflow skill:
- topic-selection → market-research
- market-research → product-design
- product-design → prd-writing
- prd-writing → system-design, ui-ux-design (parallel)
- system-design + ui-ux-design → frontend-dev, backend-dev (parallel)
- frontend-dev + backend-dev → qa-testing
- qa-testing → delivery
- delivery → COMPLETE
Update currentPhase and set next phase(s) to in_progress
Update updatedAt
Write state file
Proceed to Execute Phase for the next phase(s)

REJECT — Send Back for Revision

Trigger: $ARGUMENTS starts with reject

Read state file
Extract feedback from arguments (everything after "reject ")
Find the current phase
Set status to rejected, set rejectedAt to now, set rejectionReason to the feedback 4b. Set mode: "manual" in the state file
Update updatedAt
Write state file
Re-execute the phase (proceed to Execute Phase with the rejection feedback appended to the agent prompt)

REDO — Restart from a Specific Phase

Trigger: $ARGUMENTS starts with redo

Read state file
Extract the phase ID from arguments
Find that phase and all subsequent phases in the registry
Reset all of them to pending (clear output, approvedBy, approvedAt, rejectionReason, notes)
Set currentPhase to the specified phase
Update updatedAt
Write state file
Proceed to Execute Phase 8b. If $ARGUMENTS contains --auto: set mode: "auto" in the state file

SKIP — Skip a Phase

Trigger: $ARGUMENTS starts with skip

Read state file
Extract the phase ID from arguments
Set that phase to approved with notes: "Skipped by user"
Determine next phase and advance
Update updatedAt
Write state file
Proceed to Execute Phase

Execute Phase

This is the core orchestration loop, called by all subcommands that advance the workflow.

Sequential Phase (single phase):

Read the workflow skill: .claude/skills/workflow/SKILL.md
Read the phase skill: .claude/skills/[phase-id]/SKILL.md
Read all input documents (prior approved phase outputs)
Construct the agent prompt using the template from the workflow skill

Spawn a general-purpose Agent:

Agent({
  subagent_type: "general-purpose",
  prompt: "[constructed prompt]",
  mode: "bypassPermissions"
})

Wait for the agent to complete
Verify the output file exists (for document-producing phases)
Update state: mark phase as completed, set output path
Check mode:
- If mode is "auto": auto-approve and continue (see Auto-Pilot Protocol in workflow skill)
- If mode is "manual": display the Phase Gate message and STOP — wait for human

Parallel Phases (two agents simultaneously):

Read both phase skills
Read all input documents for both
Construct prompts for both agents

Spawn two agents in parallel (both in the same response turn):

Agent({ subagent_type: "general-purpose", prompt: "[agent-1-prompt]", mode: "bypassPermissions", run_in_background: true })
Agent({ subagent_type: "general-purpose", prompt: "[agent-2-prompt]", mode: "bypassPermissions", run_in_background: true })

Wait for both to complete
Update state: mark both phases as completed
Check mode:
- If mode is "auto": auto-approve both and continue
- If mode is "manual": display the Phase Gate message for the parallel group and STOP

Development Phases (code output, no document):

For frontend-dev and backend-dev, the agents write running code instead of documents.

Verify the code runs (check for build errors)
No output file path in state

QA Testing with Prior Rejection:

If qa-testing was previously rejected and is being re-run:

The scheduling algorithm resets frontend-dev and backend-dev to pending
Both are re-spawned with the rejection feedback
After both complete, QA testing runs again

Key Rules

Respect the mode — if mode is "manual", always stop after a phase completes. If "auto", auto-approve and continue.
State file is truth — read it at the start of every invocation, write it after every change
Use exact file paths — output files go to docs/workflow/XX-name.md as defined in the phase registry
Parallel agents run simultaneously — spawn both in the same response turn with run_in_background: true
Session recovery — if the state file exists, resume from it. Never assume state from context.
Communicate clearly — display structured gate messages, status tables, and next-step instructions

Important Notes

All 10 phases must be represented in the state file, even if some are skipped
The orchestrator itself does not produce analysis — it delegates to agents and manages state
Human decisions are final — approve means proceed, reject means redo with feedback
The workflow is linear with two parallel junctions (phases 5+6 and phases 7a+7b)

Contents

View Original Source

Related Skills

General

PromptBeginner5 minmarkdown

Untitled Skill

193

Jan 12, 2026

General

PromptBeginner5 minmarkdown

Frontend Typescript Linting.mdc

TypeScript and ESLint rules that MUST be followed when creating, modifying, or reviewing any file under apps/frontend/, including .ts, .tsx, .js, and .jsx files. Also apply when discussing frontend li...

160

Feb 15, 2026

General

PromptBeginner5 minmarkdown

2. Apply Deepthink Protocol (reason about dependencies

risks

127

Jan 15, 2026