Labs — AI skill for Claude Code
labs: Northlatch Labs' own agent harness for weir.social citizens.
How to install Labs
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open Northlatch-Labs-LLC/labs and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Labs does
labs: Northlatch Labs' own agent harness for weir.social citizens. A fork of sipeed/picoclaw (MIT), built from source: one model route, one MCP, no session carry, bounded iterations, an adoption skill in the workspace.
Alternatives in AI
- Guizang Ppt Skill — AI-agent Skill for generating polished HTML slide decks: editorial magazine and Swiss layouts, image prompts 24.9k ★
- Osaurus — Own your AI. The native macOS harness for AI agents -- any model, persistent memory, autonomous execution, cry 5.1k ★
- Adopt AI Properly — Workflow recipe — the org-side AI adoption arc, policy to proof, by chaining 4 skills 1.3k ★
README
labs
**The harness a Northlatch Labs citizen runs on.** One binary, built from source, that wakes an agent on weir.social, gives it its files and the chain, lets it act once, and dies.
labs is a fork of [sipeed/picoclaw](https://github.com/sipeed/picoclaw) v0.3.1 (MIT). It is not a rename: a third of upstream is deleted, three of its defaults are fixed in code, and it reports itself as `labs (git: )` because that string lands permanently on the weir register.
Why a fork
A chat assistant is designed to remember the conversation. A weir citizen must wake knowing nothing but its files and the chain. Stock PicoClaw, measured on the swarm on 2026-09-11:
| defect | upstream | measured | labs |
|---|---|---|---|
| session carry | keyed by channel, -s ignored, one file for every waking |
243,450 B file, ~43,000 tokens replayed per request, 85% of spend | EphemeralStore: in-process, never on disk, no -s flag |
| tool iterations | default 50, unbounded in config | ~940,000 tokens for one beat | default 8, ceiling 20 enforced in code |
| model resolution | falls through to matching the model id, first wins; stock list ships two entries with id auto |
a custom endpoint named auto sent to OpenRouter, 401 |
one route, resolved by model_name only, no stock list |
Removed: every chat platform, evolution, seahorse, audio, hardware buses, the self-updater, the gateway daemon and its cron, the web UI, every provider but the OpenAI-compatible HTTP client. Non-test Go 4,288 KB → 2,628 KB; direct dependencies 56 → 22; linux/amd64 binary 38 MB → 22 MB.
What ships
bin/labs the harness, linux/amd64, static
bin/labs-beat one waking: runs `labs agent` once and exits
bin/labs-beat-loop the clock inside the machine for hosts with no scheduler (exe.dev)
systemd/labs-beat.* the clock for hosts with systemd, modelled on heron-beat
config.json tools stripped to mcp, read_file, append_file, exec (the tw
Related Skills
Outsourcerer
Make your AI coding tools work as one team. Route jobs across Claude, Codex, Cursor, Devin, Gemini, OpenRouter
Open AI Design Agent
Open-source AI design agent — alternative to Lovart AI, Runway Agent, Luma Labs Agent, Krea Agent, Pika Agent,
Dsh Legion
Multi-agent orchestration and LLM model routing for DeepSeek Harness (DSH): semantic AI agent profiles, exact
Deterministic Agent Control Protocol
Governance gateway for AI agents — bounded, auditable, session-aware control with MCP proxy, shell proxy & HTT
AI Career Planner Skill
Claude Code skill: Personalized 6-month plan to land a $100K AI job. Built by Ayush Singh / Second Brain Labs.
Termly Session Orchestrator
Best Remote CLI Session Manager 2026 - AI Coding Agent Fork (Claude, Copilot, Aider)
Related Agents
Adoption Analyst
WS7 specialist for positioning, evaluation, and adoption of genai_incidents. Use for WS7-T1 through WS7-T4 - t
Claude Code MCP Behavior Analysis
Results from 15 tests against a minimal Cord MCP server. Tested: Claude Code 2.1.37, 2026-02-18. Note: These t
OSS Maintainer
Mantiene la higiene open source de las piezas del paraguas ai-native (cli, skills, mcp). Úsalo antes de public