Back to skills
SKILL.md
Project Onboard
ASecurityPrimary project onboarding for Claude Lane Stack. Diátaxis-classified LLM pack (MODULE_MAP, API_SURFACE, TEST_INDEX, TAXONOMY, RUNBOOK, Google DESIGN.md). Schemas + validation in references/. Weekly docs-maintainer. Use when: /project-onboard, онбординг, init project, bootstrap CLAUDE.md, deep onboard, wiki→llm docs in the terminal Lane Stack; `$ARGUMENTS` exactly info prints the card. Not inside Lane Pilot chats (reference only).
- 121 stars
- 0 votes
- 0 copies
- 2 views
- Added September 3, 2026
Works with
Security analysis
100/100Pro scans all 4 files and shows the line behind each finding
npx -y skills add VKirill/claude-lane-stack --skill project-onboard --agent claude-codeAre you the author of Project Onboard?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/vkirill-project-onboard)---
name: project-onboard
description: Primary project onboarding for Claude Lane Stack. Diátaxis-classified LLM pack (MODULE_MAP, API_SURFACE, TEST_INDEX, TAXONOMY, RUNBOOK, Google DESIGN.md). Schemas + validation in references/. Weekly docs-maintainer. Use when: /project-onboard, онбординг, init project, bootstrap CLAUDE.md, deep onboard, wiki→llm docs in the terminal Lane Stack; `$ARGUMENTS` exactly info prints the card. Not inside Lane Pilot chats (reference only).
argument-hint: "[info]"
user-invocable: false
---
# Project onboard (Claude Lane Stack)
> **Inside a Lane Pilot chat** (`LANE_PILOT_AGENT_TYPE` set / tools `lane_pilot_*` present) this skill is reference only. Do not generate `docs/llm`, and do not write `docs/`, `CLAUDE.md`, `AGENTS.md` or `llms.txt` unless the owner asked for that in this chat (Lane Pilot's nightly docs pass owns `docs/` and builds `docs/index.md`; the onboarding preview accepts only `docs/**.md` and `apps/**.md`). Orient the owner from `PROJECT.md` → `docs/index.md` and answer in Russian. The pipeline below is the terminal Lane Stack harness.
## Info
If `$ARGUMENTS` is exactly `info`, print `references/info.md` verbatim (Russian), then stop. Do not start onboard.
## Who runs it
| Role | How |
|------|-----|
| **One-shot CLI** | `project-onboard .` (detect + seed + Codex/Cursor fill) |
| **Default writer** | **Codex** (`stages.onboard` from adoc) |
| Instructions | `~/.agents/codex/instructions/onboard.md` |
| Model / effort / fast | **adoc** `stages.onboard` |
| Weekly refresh | `docs-maintainer` (`ONBOARD_REFRESH=weekly`) |
| Existing UI, no DESIGN.md | `design-lead` (skill `project-design`) — extract, do not re-onboard |
Default CLI = full pipeline. `--seed-only` = stubs without model. Do **not** use Grok.
## References (load before filling)
| Ref | What |
|-----|------|
| [references/PACK-SCHEMAS.md](references/PACK-SCHEMAS.md) | Author every file (Diátaxis + schemas) |
| [references/VALIDATION.md](references/VALIDATION.md) | Shell + acceptance; refuse `complete` |
| [references/design-md-standard.md](references/design-md-standard.md) | Full `@google/design.md` canon |
| https://diataxis.fr/ | Tutorial / how-to / reference / explanation |
| https://agents.md/ · https://llmstxt.org/ | Agent entry + curated index |
**Classifier file in every project:** `docs/llm/TAXONOMY.yaml`
We **omit tutorials** from the LLM pack.
## Pack summary
| File | Diátaxis | Load |
|------|----------|------|
| CLAUDE / AGENTS | how-to | always |
| `apps/*/CLAUDE.md` + `apps/*/docs/` | how-to + reference | on_demand (per-app pack, phase4b) |
| llms.txt / INDEX | reference | always |
| TAXONOMY / MODULE_MAP / API_SURFACE / TEST_INDEX | reference | on_demand |
| ARCHITECTURE / FLOWS / gotchas / ADR | explanation | on_demand |
| RUNBOOK / TESTING / deployment | how-to | on_demand |
| DESIGN (`has_ui`) | reference | on_demand |
## Deep pipeline
```text
Phase0 seed
→ Pass A (model): Phase1–4 root pack
→ Host: onboard_app_packs prepare (skeletons + project API_SURFACE → apps/*)
→ Pass B (model): Phase4b fill each apps/* + Phase5
→ Host: VALIDATION gate (fail thin/catch-all packs)
→ report
```
Deep monorepo (≥2 apps) uses this **two-pass pipeline**. Toy/single-app stays single-pass.
## Signals
| Signal | Extra files |
|--------|-------------|
| `has_ui` | `docs/DESIGN.md` (Google format) |
| `has_deploy` / `signals.deploy` | `docs/RUNBOOK.md` |
## Procedure (PM, terminal)
1. Passport thin: spawn **project-onboarder** (or `project-onboard`). Wait `DONE`.
2. If `stages.docs.enabled`: then spawn **docs-maintainer**. Never both at once.
3. Writer reads skill `references/`.
4. After: RU summary — surfaces, modules, tests, DESIGN?, RUNBOOK?, validation.
## After checklist
- [ ] PACK-SCHEMAS followed
- [ ] VALIDATION green or GAPS listed
- [ ] TAXONOMY / API_SURFACE / MODULE_MAP / TEST_INDEX real
- [ ] MODULE_MAP covers apps (+ runners/hot libs) or GAPS
- [ ] Phase4b: each `apps/*` walked → **filled** local CLAUDE + docs/ (not stubs; surfaces projected/split)
- [ ] Webhooks split per provider; auth unknown ≤40%
- [ ] MANIFEST always_load lean (CLAUDE/AGENTS/INDEX only)
- [ ] decisions.md 3–5 evidenced ADRs (deep full) or GAPS
- [ ] FLOWS notification includes stream handler + queue consumer when present
- [ ] has_ui ⇒ DESIGN Google format
- [ ] has_deploy ⇒ RUNBOOK
- [ ] phase artifacts + report under `.agents/runs/_onboard/artifacts/001`
- [ ] MANIFEST weekly refresh
Files in this skill
- SKILL.md
- references/PACK-SCHEMAS.md
- references/VALIDATION.md
- references/design-md-standard.md
Attribution
Comments
Loading comments…
Newsletter
Get the weekly list of safe new skills, and the ones we pulled.