onboard

name	onboard
description	First-time onboarding for a workspace that already has Workato projects and custom connectors. Thin orchestrator that pulls every project and runs the existing learn / sync / catalog skills to bootstrap the org knowledge base. Japanese prompts are also supported.
disable-model-invocation	true

/onboard

Bootstrap the organization knowledge base from Workato assets that already exist before the kit was adopted.

Most teams adopt this kit when their Workato workspace is already full of projects and custom connectors. /onboard pulls all of that down and runs the existing learn / sync / catalog skills over it, so docs/ and org/docs/ start out populated with the patterns, built-in actions, schemas and connector details this organization actually uses — making every later /create-recipe and /plan sharper.

This skill is a thin orchestrator. It does not pull, learn, sync or catalog anything itself — it sequences the existing skills and never reimplements their logic.

Usage

• /onboard — onboard every remote project and connector
• /onboard --projects "<a>","<b>" — limit to specific projects
• /onboard --resume — continue an interrupted run from the progress report
• /onboard --dry-run — show the plan and scope, execute nothing

Skills it orchestrates

Step	Skill invoked	Writes to
Pull	/pull-project --all (or --projects)	projects/
Connectors	/sync-connectors --all	docs/connectors/, connectors/docs/
Learn recipes	/learn-recipe <project> (per project)	org/docs/
Patterns	/learn-pattern	org/docs/patterns/recipe-patterns/
Catalog	/catalog scan	projects/CATALOG.md

/onboard only invokes these — see each skill for its own behaviour.

Progress report

/onboard keeps a single artifact, org/onboarding-report.md, that doubles as the resume checkpoint and the final summary. It is a checklist of steps and per-project status. --resume reads it and skips everything already marked [x]; a fresh run creates it.

# Onboarding report
Started: <YYYY-MM-DD HH:MM>   Last updated: <YYYY-MM-DD HH:MM>
Scope: all remote projects   (or: projects a, b, c)

## Steps
- [x] Pull projects        (12/12)
- [x] Sync connectors      (8 pre-built, 2 custom)
- [ ] Learn recipes        (7/12 projects)
- [ ] Learn patterns
- [ ] Catalog

## Per-project (learn-recipe)
- [x] Sales - Common
- [ ] [App] IT Onboarding   ← resume here
...

Procedure

1. Preconditions

• Confirm the Workato CLI is authenticated: workato projects list --source remote --output-mode json succeeds. If not, tell the user to run workato init and stop.
• --resume: read org/onboarding-report.md. If it does not exist, say so and start a fresh run.

2. Plan and confirm scope

• List remote projects (workato projects list --source remote --output-mode json).
• Resolve the scope: all projects, or the --projects subset.
• Estimate size and warn for large workspaces: hundreds of projects/recipes mean a long run and many API calls. Recommend --projects for a first pass, and remind the user the run is resumable.

Onboarding plan:
  Projects to pull:   <N>
  Connectors:         /sync-connectors --all
  Then per project:   /learn-recipe → /learn-pattern → /catalog scan

This may take a while on a large workspace and is rate-limited by Workato.
The run is resumable with /onboard --resume. Proceed? (Y/N)

With --dry-run, stop here after printing the plan.

3. Create / update the progress report

Create org/onboarding-report.md (or load it for --resume). Mark steps [ ].

4. Step — Pull projects

Invoke /pull-project --all — or /pull-project --projects "<a>","<b>" with the in-scope names for a --projects run. /pull-project handles both not-present (workato init) and present projects, and places each project's .workatoignore. Update the report: Pull projects (<done>/<total>).

5. Step — Sync connectors

Invoke /sync-connectors --all — pre-built connectors via API into docs/connectors/, custom connectors parsed from connector.rb into connectors/docs/. Mark the step done with counts.

6. Step — Learn recipes (per project)

For each in-scope project, invoke /learn-recipe <project-name>. /learn-recipe appends incrementally to org/docs/ and is safe to re-run.

• On a large workspace, dispatch each project's /learn-recipe as a separate Agent call so one project's recipe volume does not exhaust context.
• After each project completes, check it off in the report's per-project list. This is the resume granularity — an interrupted run continues at the first unchecked project.

7. Step — Learn patterns

Across the pulled recipes, identify recurring constructions (pagination loops, approval flows, error handlers, …) and invoke /learn-pattern for each. Skip constructs already present in docs/patterns/recipe-patterns/ or org/docs/patterns/recipe-patterns/ — learn only the delta.

8. Step — Catalog

Invoke /catalog scan to inventory shared Recipe Functions and connections into projects/CATALOG.md.

9. Summary

Finalize org/onboarding-report.md and print the summary:

Onboarding complete.

Collected:
- Projects pulled:        <N>
- Connectors documented:  <P> pre-built, <C> custom
- Recipes learned:        <R> across <N> projects
- Patterns recorded:      <K>

Most-used built-in actions (top 10):
  <provider>/<action>   ×<count>
  ...

Needs follow-up:
- Unlearned actions:      <list from /learn-recipe "Unlearned Actions">
- Projects skipped / failed: <list>

Suggested next step:
- /auto-learn <connector>   to deep-dive a heavily-used connector via the UI

Compute "most-used built-in actions" by tallying provider / action keywords across the pulled *.recipe.json files (a read-only scan — no new knowledge logic).

Scale and resume

• Resumable: every step and every project is checked off in org/onboarding-report.md. /onboard --resume re-reads it and continues at the first unchecked item. A run interrupted by a timeout, rate limit or context limit loses no completed work.
• Idempotent: re-running a completed step is safe — /pull-project respects .workatoignore, /sync-connectors and /learn-recipe append/update rather than duplicate, /catalog scan regenerates. Re-running /onboard without --resume re-verifies each step.
• Rate limits: --all issues many API calls. If the CLI starts failing, stop, let the user wait, and resume with --resume.
• Context: dispatch per-project /learn-recipe as Agent calls (Step 6) so a large project does not blow the orchestrator's context.

Editor support

/onboard is CLI / API only — it does not require a browser, so it runs in every editor (Claude Code, Cursor, Codex CLI, Gemini CLI). Browser-based deep collection is intentionally out of scope; the summary suggests /auto-learn (Claude Code + Chrome extension) as an optional follow-up for heavily-used connectors.

Rules to follow

• Don't reimplement: /onboard never pulls, learns, syncs or catalogs by itself — it only invokes the owning skills. The single piece of state it owns is org/onboarding-report.md.
• Check off reliably: mark a step / project [x] only after the invoked skill actually succeeds. On failure, record it under "Needs follow-up" and continue.
• Confirm before a large run: always show the plan and project count first; recommend --projects for a first pass.
• Never touch the kit's docs/ destructively: learning writes go to org/docs/ via /learn-recipe (see .cursor/rules/org-knowledge-overlay.mdc).

Git management

Onboarding produces a lot of files (projects/, org/docs/, docs/connectors/, CATALOG.md). Commit in the workspace repository once the user has reviewed:

git add projects/ org/ docs/ connectors/ AGENTS.md
git commit -m "onboard: bootstrap knowledge base from existing Workato assets"
git push origin

Connector docs under the kit's docs/connectors/ are commits into the kit submodule — PR them back to workato-dev-kit separately (see /sync-connectors).