LazyBrain — Codex Desktop Capability Router

Ask naturally in Codex Desktop. LazyBrain searches your local Skills, Plugins, MCP servers, agents, and commands, then recommends what to use, why, and—when useful—turns the choice into an interactive @Visualize decision explorer.

LazyBrain is a Codex Desktop-first, local capability index and deterministic router. It inventories local Skills, Plugins, MCP server metadata, agents, slash commands, rules, and workflow templates, then turns a plain-language task—even a vague one—into an explainable choice, comparison, clarification, or execution order.

It is built for developers who already have powerful agent tools installed and do not want to remember every exact command.

npm install -g lazybrain
lb quickstart
lb ask "review this payment PR safely"

Why LazyBrain

Problem	LazyBrain gives you
Too many skills, commands, and rules to remember	One natural-language entrypoint
Agents picking random tools or generic workflows	Deterministic local routing
A vague prompt that could match several tools	One recommendation, tradeoffs, or a clarification question
Repeated release, security, migration, and review workflows	Reusable combos and orchestration plans
Hook suggestions that interrupt too often	Quiet suggestions only when confidence is high
Concern about scanned files leaving the machine	Local graph, local cache, no telemetry

What It Does

Surface	Status	Use it for
Codex Desktop plugin + bundled Skill	Ready for local checkout	Ask in the conversation, search local capabilities, compare choices, and use @Visualize when available
lb CLI	Support surface	Initialize the local graph, inspect decisions, and debug the desktop backend
Claude Code hook	Ready	Automatic high-confidence suggestions inside a project
lazybrain-mcp	Ready	Deterministic routing from MCP-capable agent clients
Local graph/cache	Ready	Fast matching from local capability metadata
Hosted dashboard	Not included	No cloud UI or team sync in this beta
Automatic task execution	Not included	LazyBrain recommends and plans; your agent executes

Current version: 2.1.0.

Install

Requires Node.js 18 or newer.

npm install -g lazybrain
lb quickstart
lb ready

npm install only installs the CLI. It does not scan your home directory. lb quickstart is the explicit first-run command that scans local capability metadata and builds ~/.lazybrain/graph.json.

Beta channel:

npm install -g lazybrain@beta

GitHub release tarball after v2.1.0 is published:

npm install -g https://github.com/papperrollinggery/lazy-brain/releases/download/v2.1.0/lazybrain-2.1.0.tgz

Source checkout:

git clone https://github.com/papperrollinggery/lazy-brain.git
cd lazy-brain
npm ci
npm run build
npm link
lb quickstart
lb ready

Full install, cleanup, MCP, and smoke-test instructions: docs/INSTALL.md.

Codex Desktop quick start

The source checkout contains the Codex plugin, Skill, MCP declaration, and local marketplace entry:

npm ci
npm run build
npm link
codex plugin marketplace add .
codex plugin add lazybrain@lazybrain-local

Start a new task in Codex Desktop and ask which installed capability best fits your goal. LazyBrain calls its read-only recommendation tool. For an interactive comparison, explicitly select @Visualize in the composer before sending the task. When the result contains desktopVisualization.shouldRender: true and @Visualize is exposed in that task, the Skill passes its exact visualization prompt to it. If the preview is unavailable or was not selected, the task receives an accessible Markdown table and a ready-to-reuse visualization prompt. See Codex Desktop integration.

Quick Demo

Find the right capability:

lb "review this PR for security issues"

Turn a vague prompt into a structured decision for Codex, Claude Code, or another client:

lb ask "help me ship this safely" --json

Inspect the exact Codex Desktop visualization contract:

lb desktop "review this payment PR safely" --json
lb desktop "review this payment PR safely" --visualize-prompt

The decision contract is fail-closed: low-confidence prompts return clarify instead of silently choosing a tool.

Example output:

/security-review 98%
Scan code for OWASP Top 10, auth bypass, injection, and credential exposure.

Also consider:
- /code-review
- /gitnexus-pr-review

Turn a risky task into an ordered plan:

lb orchestrate "deploy payment feature"

Pick a reusable workflow:

lb combo "deploy new feature to production"

Install quiet Claude Code suggestions for the current project:

lb hook install
lb hook status

After the hook is installed, you can keep typing normal prompts in Claude Code. LazyBrain only adds a suggestion when the match is confident.

Commands

Command	Purpose
lb "task"	Find the best matching capability
lb ask "task"	Choose, compare, or clarify; add --json for agents and visualizations
lb desktop "task"	Return the Codex Desktop @Visualize decision payload, accessible fallback, or exact visualization prompt
lb use <name> [task]	Explicitly record that you chose a recommendation
lb combo "task"	Return a reusable workflow template
lb orchestrate "task"	Build a multi-skill execution plan
lb scan	Scan local capability files
lb compile	Rebuild the local capability graph; no LLM or embedding call
lb quickstart	Scan and compile in one first-run command
lb stats	Show recent usage and patterns
lb discover	Find high-value unused local capabilities
lb config show	Print local config with secrets redacted
lb ready / lb ready --json	Check graph and hook readiness
lb hook plan	Show the hook change that would be made
lb hook install	Install the project Claude Code hook
lb hook uninstall	Remove the project hook
lazybrain-mcp	Start the stdio MCP server

MCP

Add LazyBrain to an MCP-capable client:

{
  "mcpServers": {
    "lazybrain": {
      "command": "lazybrain-mcp",
      "args": []
    }
  }
}

Source checkout variant:

{
  "mcpServers": {
    "lazybrain": {
      "command": "node",
      "args": ["/absolute/path/to/lazy-brain/dist/bin/mcp.js"]
    }
  }
}

Current MCP tools:

Tool	Purpose
lazybrain_find	Find matching capabilities for a task
lazybrain_recommend	Return a backward-compatible decision plus desktopVisualization, alternatives, reasons, and optional sequence
lazybrain_orchestrate	Build an orchestration plan
lazybrain_catalog	Summarize the indexed capability library by type
lazybrain_stats	Read recent local usage stats
lazybrain_scan	Scan local capability sources

Smoke test:

printf '{"jsonrpc":"2.0","id":1,"method":"tools/list"}\n' | lazybrain-mcp

Codex Desktop and Claude Code

Codex Desktop is the primary product surface; Claude Code remains a supported compatibility surface:

• Codex Desktop: packaged plugin manifest, bundled $lazybrain-find Skill, read-only MCP recommendation tools, and a versioned interactive-decision payload for the installed @Visualize plugin.
• The decision explorer keeps scores, reasons, source, platform, alternatives, and workflow visible; candidate selection never executes a capability.
• @Visualize is an OpenAI preview and must be selected in the composer for the task; availability may depend on account/workspace rollout. LazyBrain falls back when it is not exposed.
• Claude Code: quiet UserPromptSubmit hook plus the same deterministic core, CLI, and MCP server, without claiming Desktop visualization rendering.

Local Codex plugin development from this checkout:

npm ci
npm run build
npm link
codex plugin marketplace add .
codex plugin add lazybrain@lazybrain-local

This changes personal Codex configuration, so run it only when you want to install the local development plugin. Start a new Codex task after installation. The project does not auto-install plugins or edit Codex configuration.

What LazyBrain Indexes

lb quickstart, lb scan, and lb compile read local capability metadata from common agent-tool locations:

• Claude Code skills and commands
• Codex and universal .agents skills
• installed Codex/Claude plugin manifests and bundled skills, agents, commands, and MCP declarations
• Codex config.toml, Claude .mcp.json/configuration MCP server names (credential values are never copied into the graph)
• project .claude/commands
• .skillshub
• .codex/skills
• .agents/skills
• Cursor, Windsurf, Cline, and OpenCode rule files
• local SKILL.md-style capability files

Empty machines still work because LazyBrain includes built-in capabilities for common development workflows.

Reliability

LazyBrain's hot path is deterministic:

• no runtime LLM call for normal matching
• no embedding dependency for normal matching
• no runtime dependencies in the published package
• low-confidence decisions ask one clarifying question instead of guessing
• MCP tools declare read-only/open-world/destructive safety annotations
• low-confidence hook suggestions stay silent
• golden-set tests cover 76 labeled routing cases plus negative cases
• precision gate requires at least 88% top-match precision
• latency gate requires average find() time under 200ms

Verification commands:

npm run lint
npm test
npm run build
npm run audit:public
npm pack --dry-run --json

Privacy

LazyBrain is local-first. It scans local capability metadata and writes local cache/history files under ~/.lazybrain. It does not upload scanned files, does not require a cloud account, and does not send telemetry.

Details: docs/PRIVACY.md.

Best Fit

Use LazyBrain when you have:

• many local agent skills, slash commands, rules, prompts, or plugins
• multiple MCP servers and no reliable memory of which one fits a task
• repeated workflows such as release, security review, migration, incident response, or PR review
• an MCP-capable agent client that needs deterministic tool selection
• a preference for local routing over runtime model calls

Wait if you need:

• fully automatic task execution
• hosted team dashboards
• cross-machine sync
• managed cloud analytics

Docs

• Install
• Use cases
• Product direction and architecture
• Codex Desktop and @Visualize
• Privacy
• Release checklist

FAQ

Is LazyBrain another Skill or MCP installer?

No. Native marketplaces and registries already handle installation. LazyBrain indexes what is available locally and helps the agent choose the right capability at execution time.

Does LazyBrain send my prompts or local tool metadata to an LLM?

Normal matching is deterministic and local. LazyBrain reads capability metadata and stores its graph/history under ~/.lazybrain; it does not require an API key or telemetry service.

Can it execute the recommended workflow automatically?

No. Recommendations and orchestration plans are advisory. Codex, Claude Code, or the user decides whether to invoke a capability, especially when it may write, publish, install, or change external systems.

Why not embeddings or an LLM router?

The default path favors auditable triggers, examples, local history, and high confidence thresholds. It stays fast, offline, debuggable, and easy to improve with a golden test case.

Does LazyBrain itself render the interactive interface?

LazyBrain produces a bounded local decision snapshot and exact visualization prompt. Capability metadata is treated as untrusted display data. In Codex Desktop, the installed OpenAI @Visualize plugin renders the interactive explorer when that preview is available. LazyBrain never substitutes invented data or treats a card selection as permission to execute a tool.

Contributing

The smallest useful PR is one trigger phrase plus one golden-set case:

1. Add the trigger/example in src/knowledge/builtin.ts.
2. Add a labeled query in test/golden/find.test.ts.
3. Run npm test.

Useful contribution areas: trigger phrases, combo templates, orchestration rules, scanner coverage, and benchmark cases.

License

AGPL-3.0.