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.

Project onboard (Claude Lane Stack)

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)

Default CLI = full pipeline. --seed-only = stubs without model. Do not use Grok.

References (MUST load before filling)

Ref	What
references/PACK-SCHEMAS.md	Author every file (Diátaxis + schemas)
references/VALIDATION.md	Shell + acceptance; refuse complete
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

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

MUST (PM)

1. Prefer: project-onboard "$PROJECT_CWD" (or spawn project-onboarder).
2. Writer reads skill references/.
3. After: RU summary — surfaces, modules, tests, DESIGN?, RUNBOOK?, validation.
4. Weekly: ONBOARD_REFRESH=weekly → docs-maintainer.

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