Skip to main content
Agent skill

Lobu Operator

Contribute safely to the Lobu monorepo: worktrees, package rules, red-to-green fixes, validation gates, SDK-first operations, PRs, and rollout checks.

Get started

Get the complete skill folder, including its scripts and reference files.

When to use this skill

Use Lobu Operator when an AI agent needs a reusable SKILL.md workflow for this job: Contribute safely to the Lobu monorepo: worktrees, package rules, red-to-green fixes, validation gates, SDK-first operations, PRs, and rollout checks.

When not to use it

Skip Lobu Operator when the task is outside the coding category, or when a more specific skill in this directory already covers the same workflow with clearer triggers.

How to install

Where will you use it?

Install scope
Claude Code folder setup
mkdir -p ~/.claude/skills/lobu-operator
# Save SKILL.md to ~/.claude/skills/lobu-operator/SKILL.md
  1. Claude Code personal install: place the skill and its bundled files at ~/.claude/skills/lobu-operator/SKILL.md. For a shared project, commit the folder at .claude/skills/lobu-operator/ instead.
  2. Current Claude Code detects SKILL.md changes in watched directories during the session. If you just created a top-level skills directory, run /reload-skills. Check that the skill is listed before testing its trigger.
  3. For Claude chat or Cowork in the Desktop app, manage uploaded skills in Customize > Skills; copying files into a local Claude Code folder is not a Desktop install. Other agents may use different install locations.

Full install guide for Claude, Cursor, and Codex →

Try a starter prompt

After installing the skill, paste this starter prompt into your agent to verify it triggers on a small task.

Starter prompt
Use the Lobu Operator skill for this task: Use Lobu Operator when this documented workflow matches the task. Confirm any required inputs or files before making changes.

What supports this recommendation?

70/100 provisional content score. This measures documentation, practical detail and source signals. It is not a measured task success rate or a security certification.

Documentation
76/100
Practical detail
68/100
Source evidence
62/100
Source signals
72/100
Model compatibility evidence
  • claudeinferred

    The skill text mentions Claude or a closely associated term. This is keyword-derived compatibility; no hands-on model test is recorded.

  • chatgptuntested

    No model-specific signal or recorded compatibility test was found.

  • geminiuntested

    No model-specific signal or recorded compatibility test was found.

  • copilotuntested

    No model-specific signal or recorded compatibility test was found.

  • llamauntested

    No model-specific signal or recorded compatibility test was found.

  • perplexityuntested

    No model-specific signal or recorded compatibility test was found.

  • mistraluntested

    No model-specific signal or recorded compatibility test was found.

  • grokuntested

    No model-specific signal or recorded compatibility test was found.

Inferred compatibility is based on content signals. Check the app, its available tools and setup requirements before running a skill.

What this skill does

# Lobu Operator — Repo Guide

This is a fast index, not a replacement for repository instructions. Root `AGENTS.md` and the nearest package `AGENTS.md` are authoritative; `CLAUDE.md` includes the root rules for Claude sessions.

## Before You Act

1. Read root `AGENTS.md`, the touched package's nearest `AGENTS.md`, and the relevant `docs/GOTCHAS.md` section. 2. Read the concept docs before touching automation: `docs/CONCEPTS.md` (entities vs events, identity, end-to-end lifecycle), `docs/AUTOMATIONS.md` (Automation contract), `docs/connector-authoring.md` (custom connectors), `docs/README.md` (index). 3. Run `make task-setup NAME=<slug>` and work only in the resulting `.claude/worktrees/<slug>/` directory. Never switch branches or edit in the main checkout. 4. Read `lobu.config.ts` when configuration or runtime semantics matters; inspect the active agent and skill directories because composition is data-driven. 5. Reproduce a bug before changing code. Capture red→fix→green evidence and exercise every branch touched; a typecheck alone is not completion.

## Dev Workflow

```bash make task-setup NAME=<slug> cd .claude/worktrees/<slug> make dev # gateway + workers + Vite on the allocated port make clean-workers # reap an orphaned gateway from a crashed dev run ```

Prerequisites are Bun, the supported Node version, and Postgres with pgvector via `DATABASE_URL`. `make dev` uses the shared local Postgres; `make dev-embedded` uses per-worktree embedded Postgres. Read `.env.local` for the allocated ports.

## Correctness Invariants

- Design for at least three replicas. Shared required state and cross-pod signals belong in Postgres, never a process-local singleton or `Map`. - `events` is append-only. Replace or hide records with superseding/tombstone events; never delete history. - User-facing reads must not aggregate growing history. Materialize bounded answers on writes and read them back by index. - Never bulk-delete production organizations. Treat apparently empty organizations as real signups requiring individual confirmation. - Workers receive placeholders/proxied access, device-pinned credentials, or short-lived provider-derived leases—never durable stored credentials. - Agent-facing vocabulary always says **Automation**; engine-only vocabulary stays internal. - Durable dispatch and delivery failures fail closed. Retry, defer, or surface terminal failure; never reinterpret an ambiguous coordination error as permission to proceed.

## Validate and Ship

Run focused tests while iterating, then the settled-diff gates in this order:

```bash bun test <path> # focused local iteration git add -- <paths> # explicit paths only, never -A make pre-pr # fast local gates before commit/push git commit -m '<type>(<scope>): <summary>' git diff --name-only origin/main...HEAD git push -u origin <branch> gh pr create make review # bounded review/fix/independent verification make ui-review make land N=<number> ```

GitHub CI (`ci.yml`) is the canonical gate: free on this public repo, full Linux graph in ~5–7 min per PR. `make pre-pr` runs the fast local gates (typecheck, knip, lint, naming) before push; `make review` can run alongside CI and publishes success only after exact-HEAD CI passes. Optional findings never reopen a pass; the branch has two automatic write attempts across all invocations/resumes. `make pr-full` (Daytona ephemeral sandbox, else local) is optional tooling, not part of the required loop. Stage every intended new file explicitly before pushing.

Never bypass a check that has not reported. For a production-visible change, wait for deployment and prove the PR's squash merge commit is an ancestor of the deployed SHA before running the live check. Clean up the task worktree with `make task-clean` after merge.

## Data Integration & Knowledge Ingestion

- Discover the current ClientSDK with `search_sdk`; use `query_sdk` for reads and `run_sdk` / `lobu call run_sdk` (with a `title` arg) for writes. - Connectors plus feeds are the normal integration path. Use `connections.connect`, follow any `setup_required` continuation, then create and trigger a feed with `feeds.create` and `feeds.trigger`. - Find connector actions with `operations.listAvailable` and execute the returned target with `operations.execute`; do not guess operation or connection identifiers. - Use `knowledge.save` for schema-less semantic history and `entities.create` / `entities.update` for strict structured records. Chunk bulk work and use `Promise.allSettled` so conflicts are explicit. - `lobu memory seed` is suitable for small declarative YAML datasets, not large backfills.

Intended uses

  • Use Lobu Operator when this documented workflow matches the task.

Implementation notes

  • Review the source instructions and adapt inputs before running the workflow.

Try it, then tell us how it went

A copied prompt does not tell us whether the task worked. Report your result after checking the output. We collect your tool and outcome, never your files or prompt.