LazyBrain

Type a task once. LazyBrain finds the right local AI capability and turns it into a deterministic orchestration plan.

The Problem

You installed hundreds of AI skills, commands, plugins, and local rules. You remember the same twenty. LazyBrain makes the rest usable from plain language.

Install

Current beta prerelease:

npm install -g https://github.com/papperrollinggery/lazy-brain/releases/download/v2.0.0-beta.1/lazybrain-2.0.0.tgz
lb quickstart
lb "review this PR for security issues"

Registry beta, after npm publish:

npm install -g lazybrain@beta

From source:

npm ci
npm run build
node dist/bin/lazybrain.js quickstart

See docs/INSTALL.md for MCP setup, local capability paths, and source checkout usage.

What It Does

lb "review this PR for security issues"     # best matching capability
lb combo "deploy new feature to production" # reusable workflow template
lb orchestrate "deploy payment feature"     # multi-skill execution plan
lb stats                                    # recent usage and patterns
lb discover                                 # high-value unused capabilities
lb scan && lb compile                       # refresh the local knowledge graph
lb config show                              # inspect redacted local config
lazybrain-mcp                               # start the stdio MCP server

How It Works

task text
  -> trigger, tag, and example matcher
  -> optional local graph from scan/compile
  -> combo registry and orchestration rules
  -> CLI, hook, and statusline surfaces

The hot path is deterministic: no runtime LLM call, no embedding dependency for matching, and low-confidence hook cases stay silent.

Works With

LazyBrain scans or indexes local capability files for:

Claude Code Codex Cursor Windsurf Cline OpenCode local SKILL.md

Default scan paths include Claude skills/commands, project commands, Cursor/Windsurf/Cline rule files, .skillshub, .codex/skills, and .agents/skills.

LazyBrain is local-first. It scans local capability metadata and writes local cache/history files; it does not call an LLM on the hot path and does not upload scanned files. See docs/PRIVACY.md.

Orchestration

lb orchestrate upgrades a single task into an ordered work plan:

$ lb orchestrate "deploy payment feature"

Orchestration Plan 94%
payment/auth risk detected

1. /security-review
2. /tdd-workflow
3. /code-review
4. /ship

Sequence: sequential
Auto-activate: no

Workflow templates are available through lb combo:

$ lb combo "deploy new feature to production"

Recommended workflow: release_public_audit
1. /document-release
2. /github-ops
3. /ci-cd-best-practices

Verification: npm run audit:public && npm pack --dry-run --json

MCP Server

Installed package:

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

Source checkout:

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

Benchmarks

Repository-backed checks:

Claim	Evidence
Golden-set routing	76 labeled cases plus negative checks in test/golden/find.test.ts
Precision gate	test requires at least 88% exact top-match precision
Latency gate	test requires average find() latency under 200ms across 100 runs
Built-in matcher surface	core skills plus generated capability names in src/knowledge/builtin.ts
Orchestration surface	18 rules in src/orchestrator/rules.ts, 12 combos in src/combos/registry.ts
Runtime model	deterministic matcher and rule engine; no runtime LLM call on the hot path

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.

User Scenarios

See docs/USE_CASES.md for supported beta workflows and examples.

Release Readiness

See docs/RELEASE_CHECKLIST.md before publishing a beta or stable release.

License

AGPL-3.0.