Menu
Plain-English pipeline from idea to production — routes user requests to the right phase and agent
The review-first prompt is selected by default. You can switch to a direct command or download a local copy.
Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
npx skills add https://github.com/CleanExpo/Unite-Group --skill idea-to-productionThe command stays on one line. Scroll horizontally to inspect it before copying.
Prefer a local copy? Download the files currently available to SkillsMP.
The authoritative environment-variable registry for the Unite-Group Nexus. Use whenever adding, reading, renaming, or debugging an env var or secret — new integration config, a `process.env.X` read, an auth/OAuth wiring, a cron secret, a "which variable holds this" question, or a "works locally but not in prod" symptom. Also use before pinning any variable name in code, and when reconciling the two credential planes (Vercel prod vs the local hermes fleet). Prevents the `APIFY_API_KEY` vs `APIFY_API_TOKEN` class of drift.
Create, verify, and promote Supabase schema changes via database branches. Every schema change/migration must be validated on a Supabase database branch before prod (see CLAUDE.md line 44).
Verify production Supabase schema before shipping code that touches it. Use BEFORE writing or reviewing ANY code that reads or writes a Supabase table in the Nexus — new queries, inserts, updates, CHECK-constrained values, RLS-dependent reads, cron data access — even one-line changes and even when the table "obviously" exists. Also use immediately when a query fails with "column does not exist", when inserts appear to succeed but produce no rows, or when generated TypeScript types disagree with runtime behaviour.
Showing SKILL.md
Based on SOC occupation classification
| name | idea-to-production |
| version | 1.0.0 |
| description | Plain-English pipeline from idea to production — routes user requests to the right phase and agent |
| triggers | ["idea","build me","i want to","let's create","let's build","can we add","from scratch","new feature","full pipeline","end to end"] |
You have an idea. Here is exactly what happens next, who does the work, and when you need to step in.
When you say "build me X" or "I have an idea for Y", Claude doesn't just start writing code. It runs your request through an 8-phase pipeline that takes you from raw idea to a reviewed, tested, production-ready pull request — with a human review gate at the end before anything ships.
You don't need to know the phases. Claude routes everything automatically. This skill documents how that routing works so you always know where you are.
| Phase | Name | What Happens | Who Does It |
|---|---|---|---|
| 1 | Intake | Claude reads your request, classifies what kind of task it is, and decides how much process to apply | Orchestrator |
| 2 | Discovery | A Product Strategist agent writes a one-page spec (PRD) that defines the problem, who it is for, and what is in scope | product-strategist |
| 3 | Decomposition | A Technical Architect maps out what files, APIs, and database tables need to change. A Senior Engineer breaks that into tasks | technical-architect, senior-engineer |
| 4 | Execution | Specialist agents build the feature in parallel, writing tests first (TDD), then code | Frontend, backend, database specialists |
| 5 | Aggregation | The Orchestrator collects all the work, checks nothing conflicts, and runs the full test suite | Orchestrator |
| 6 | Verification | Three independent checks run in parallel: code quality, acceptance criteria, and design review | verification, qa-validator, design-reviewer |
| 7 | Iteration | If anything fails verification, the responsible agent fixes it. Maximum 2 fix cycles | Specialists |
| 8 | Production | A PR is created with a full evidence trail. The pipeline stops here — you review and merge | delivery-manager then you |
Claude listens to your natural language and routes to the right starting phase.
| What You Say | What Claude Hears | Starting Phase |
|---|---|---|
| "I have an idea for..." | New product or feature concept | Phase 1 — full pipeline |
| "Build me X" | Feature request, scope unknown | Phase 1 — full pipeline |
| "Let's create a..." | New thing from scratch | Phase 1 — full pipeline |
| "Can we add X to the existing Y?" | Feature addition, codebase exists | Phase 1 — full pipeline |
| "Make the button say Submit instead of Send" | Copy change, trivial edit | Phase 4 — skip discovery |
| "Fix the login page" | Bug fix | Phase 4 — root-cause first |
| "Make it work" | Something is broken | Phase 4 — diagnose then fix |
| "Is it ready?" | Needs verification | Phase 6 — verification only |
| "Ship it" | Ready to release | Phase 6 then 8 — verify then PR |
| "What does this do?" | Explanation, no build needed | No pipeline — answer only |
Before running the pipeline, Claude answers these three questions to decide how much process to apply.
Question 1: Is this a single, isolated change?
Question 2: Does this require changes to the database, API contracts, or authentication?
Question 3: Could this break existing users if it goes wrong?
| Scope | Examples | Phases Used |
|---|---|---|
| Trivial | Change a label, tweak a config value, fix a typo | 4 then 6 then 8 |
| Standard | New page, new API endpoint, new feature flag | 1 through 8 |
| Complex | New module, database migration, auth changes, new system | 1 through 8 extended |
"I have a new idea"
Describe your idea in plain English in chat. Claude classifies scope automatically.
For complex features, it will ask up to 3 scoping questions before starting.
To explicitly trigger the discovery phase: /new-feature "your idea description"
"Fix something that is broken"
Describe the symptom, not the cause. Claude runs systematic-debugging first, then implements the fix.
pnpm turbo run test # Check what is failing before Claude starts
"Check if it is ready to ship"
pnpm turbo run type-check # Types
pnpm turbo run lint # Code style
pnpm turbo run test # All tests
Claude also runs a design review if any frontend files were changed.
"Create a PR"
Claude creates the PR at Phase 8 with full evidence. You review and merge. Claude never merges automatically.
"Something went wrong in the pipeline"
Claude will escalate to you with a clear ESCALATION report — what failed, why it cannot auto-fix, and what decision you need to make.
Claude runs autonomously through the pipeline but stops for your input when:
| Signal | What Claude Does |
|---|---|
| Requirements are ambiguous after one clarifying question | Outputs BLUEPRINT_ESCALATION — describes what is unclear and what it needs from you |
| A fix attempt fails twice (Phase 7 cap reached) | Stops, reports what failed, asks you to decide the next step |
| The change touches authentication, database schema, or auth config | Pauses and confirms with you before executing — these are HIGH risk |
| Rubric score below 50 after two iteration cycles | Escalates — the output is not good enough and needs your direction |
| Multiple agents return failures simultaneously | Escalates — too many moving parts to auto-resolve safely |
When you see an escalation, read the report. It will always tell you:
The pipeline handles all of this automatically:
pnpm turbo run type-check)pnpm turbo run lint)pnpm turbo run test)Owner: orchestrator
Owner: product-strategist
Owner: technical-architect then senior-engineer
Owner: orchestrator (coordination), specialists (implementation)
Owner: orchestrator
pnpm turbo run type-check lint testThree parallel tracks — all must clear thresholds or Phase 7 runs:
| Track | Agent | Checks |
|---|---|---|
| Code | verification | type-check, lint, test |
| Acceptance | qa-validator | rubric score 0-100 |
| Design | design-reviewer | UX audit (frontend changes only) |
Owner: orchestrator
Owner: delivery-manager
| What | Where |
|---|---|
| Full harness protocol | .claude/AGENT_HARNESS.md |
| All agent definitions | .claude/agents/*/agent.md |
| Idea to PRD workflow | .claude/workflows/idea-to-prd.md |
| PRD to Spec workflow | .claude/workflows/prd-to-spec.md |
| Spec to Build workflow | .claude/workflows/spec-to-build.md |
| Build to Release workflow | .claude/workflows/build-to-release.md |
| Quality rubrics | .claude/rubrics/ |
| Orchestrator agent | .claude/agents/orchestrator/agent.md |
| All 65 skills | .skills/AGENTS.md |
You describe what you want. Claude figures out the scope, writes the spec, builds it with tests, verifies it three ways, and hands you a PR. You review and merge. Nothing ships without you.