Pipeline Onboard

---
name: pipeline-onboard
description: >
  New project onboarding pipeline: tech-stack (profile) → CLAUDE.md setup → memory init →
  skill-routing.md check. Brings a new or inherited project into the agency in one
  command. Trigger when: starting work on a project that has no ~/.claude/projects/{slug}/
  memory structure yet; taking over an existing codebase; setting up a new repo for
  PD-managed development; when /new-project alone is insufficient (it creates scaffolding
  but doesn't profile the stack, verify CLAUDE.md completeness, or link skill-routing).
  Key capabilities: scans actual project structure, generates tech-profile.md, ensures
  CLAUDE.md has required fields, initializes memory directory, seeds task-store, and
  cross-checks skill-routing.md for stack-relevant anti-redundancy rules.
  Do NOT chain tech-stack profiler separately after this pipeline — it runs inside Stage 1.
---

# Pipeline: Onboard

Bring a project into this system. One command, four stages.

**Anti-redundancy:** This pipeline calls tech-stack (profiler mode) in Stage 1.
Do NOT add a separate `/tech-stack` call after running this pipeline.

---

## Inputs

Required:
- **Project slug**: the short name used in `/pd-spawn`, `/recall`, `medium-term.md`
- **Project path**: absolute path to the project root (e.g. `~/projects/my-app`)

Optional:
- **PD needed**: `yes` if this project needs a PD registered in `medium-term.md`
- **Description**: one-line description for medium-term.md

---

## Stage 1 — Tech Stack Profile (tech-stack profiler)

Invoke `/tech-stack` in **profiler mode**:
- Project: {project-path}
- Output: `~/.claude/projects/{slug}/memory/tech-profile.md`

This scans package.json, CLAUDE.md, docker-compose.yml, deployment configs, and
source structure. Output is a structured tech-profile.md.

---

## Stage 2 — CLAUDE.md Completeness Check

Read `{project-path}/CLAUDE.md` (if it exists). Verify it has:

| Required field | Check |
|---------------|-------|
| Project overview (what it is) | present / missing |
| Dev server command | present / missing |
| Test command | present / missing |
| Build command | present / missing |
| Deploy command | present / missing |
| Directory structure | present / missing |
| Key conventions | present / missing |

For each missing field:
1. Infer from tech-profile.md (package.json scripts, etc.)
2. Add it to CLAUDE.md if confident
3. Leave a `# TODO: verify` comment if uncertain — never invent commands

If no CLAUDE.md exists: create one at `{project-path}/CLAUDE.md` with all detected
fields pre-filled from tech-profile.md. Include a header:
```
# CLAUDE.md — {project-name}
# Auto-generated by /pipeline-onboard on {date}. Verify all commands before use.
```

---

## Stage 3 — Memory Init

Create the standard memory structure at `~/.claude/projects/{slug}/memory/`:

```
memory/
  heartbeat.md          — session state, phase, blockers
  decisions.md          — locked decisions
  next-session.md       — startup briefing for PD
  tasks/
    ongoing/            — active tasks
    completed/          — finished tasks
    revisions/          — revision requests
  inter-spawn-tasks/
    incoming/           — tasks from other PDs
    completed/          — processed tasks
    index.md            — SSOT for inter-spawn state
  sessions/             — session logs (YYYY-MM-DD.md)
  lessons/              — stack-specific lessons
  qa/                   — QA reports and screenshots
  agents/               — PD + Coord scratch files
    coords/
    executors/
  brand-guidelines.md   — brand reference (stub if unknown)
  tech-profile.md       — from Stage 1
```

For any file that already exists: skip (do not overwrite).

Write `heartbeat.md`:
```markdown
# {slug} Heartbeat

## Phase Status
Phase: ONBOARDING
Next: First PD session — read next-session.md

## Blockers
none

## Session End — {date}
### Completed This Session
- Project onboarded via /pipeline-onboard

### In Progress
- none

### Top 3 Priorities
1. Run first PD session to define project roadmap
2. Verify CLAUDE.md commands are correct
3. Set up any missing environment variables
```

Write `next-session.md`:
```markdown
# {slug}
Phase: ONBOARDED — ready for first PD session
Next: Run /pd-spawn {slug} to start project work, or /recall {slug} to review setup
Blockers: none
Decisions: none
Mid-flight: none
Delegated: none
Last saved: {date}
```

Write `inter-spawn-tasks/index.md`:
```markdown
# Inter-Spawn Tasks — {slug}

## Active Summary
_(no active inter-spawn tasks)_
```

---

## Stage 4 — Skill Routing Check

Read `~/.claude/memory/skill-routing.md`.

From tech-profile.md, identify the relevant tech domains:
- If Next.js detected: note `pipeline-feature`, `pipeline-deploy`, `pipeline-content` are relevant
- If Python detected: note that different pipeline-deploy targets apply
- If mobile (React Native / Flutter): note design-review pipeline differences

Print a brief "relevant skills for this stack" summary:
```
SKILL ROUTING CHECK — {slug}
Stack: {language} + {framework} + {deployment}
Relevant pipelines: pipeline-feature, pipeline-deploy, pipeline-content
Key anti-redundancy rules:
  - Use autoplan only (not autoplan + plan-ceo-review separately)
  - Use pipeline-deploy only (not security-critique + canary alongside it)
  - Quality gate: content-critique → content-polish (not atomics)
Full rules: ~/.claude/memory/skill-routing.md
```

---

## Stage 5 — Register in medium-term.md (optional)

If PD needed = `yes`:

Read `~/.claude/memory/medium-term.md`. In the Active Projects table, add:

```
| {slug} | {description} | {project-path} | ACTIVE |
```

If the project is already in the table: skip.

---

## Completion Report

```
PIPELINE-ONBOARD COMPLETE — {slug}

Stage 1 — Tech Profile:    ~/.claude/projects/{slug}/memory/tech-profile.md
Stage 2 — CLAUDE.md:       {project-path}/CLAUDE.md ({N fields verified, M added})
Stage 3 — Memory:          ~/.claude/projects/{slug}/memory/ ({N dirs created, M skipped)
Stage 4 — Skill Routing:   {N relevant rules identified}
Stage 5 — medium-term.md:  {added / skipped / not requested}

Next steps:
1. Verify commands in {project-path}/CLAUDE.md
2. Run: /pd-spawn {slug}  (or /recall {slug} to review without spawning)
```