Skip to content

Nidvue Teamwork

Owner: Technical Writer.

Related: Installation · Quickstart · CLI reference · Architecture · CI pipeline

Nidvue Teamwork turns one prompt into a self-organising team of AI agents that plans, builds, tests and reviews a project until every acceptance criterion is verified. It is a skill for your coding agent - not a standalone app and not a hosted service. The stdlib-only engine and the workspace format ship inside the skill; installing the engine as a Python package is optional.

  • Automatic model routing - teamwork model auto detects the agent host and routes the strongest available model to the decision lanes and the economy one to the mechanical lanes: best output where it compounds, lowest cost where it does not.

What it is

Face What it means
Skill (the product) SKILL.md, roles/, references/, templates/ and the engine in one directory, cloned into the skills directory an agent host scans (project .agents/skills/teamwork/ or user ~/.agents/skills/teamwork/). Hosts without a skill loader use the same protocol through AGENTS.md. Invoked by naming the skill in the request.
Engine (optional CLI) src/teamwork/, installable as teamwork-agents, exposing the teamwork console script and python -m teamwork. No third-party runtime dependencies. The skill runs the same engine from its own directory.
Workspace format Markdown and JSON under <project>/.pi/Teamwork/Teams/<Team>/, hidden and local; project output is committed.

One engine backs all three; the skill is how people install and use it.

Why

  • Run to completion. Work ends only when every acceptance criterion has executed evidence, not when an agent says it is done.
  • Evidence over claims. A task cannot reach done without a recorded command, revision and observed result.
  • Independent review. A different role reviews every deliverable before it is merged; only the Reviewer may merge.
  • Parallel lanes. Interfaces are frozen first, then developer, tester and designer work concurrently against the same contract.
  • Auditable by design. Every mutation is appended to a hash-chained ledger (Ledger.jsonl) you can verify without trusting the team.

30-second quickstart

$ mkdir -p .agents/skills && git clone --depth 1 https://github.com/Nidvue/Teamwork.git .agents/skills/teamwork && rm -rf .agents/skills/teamwork/.git
$ cd my-project
> use the teamwork skill: build a URL shortener with a CLI, tests and docs

One clone installs the skill; the prompt starts the company. The CEO writes AGENTS.md, staffs the roster and opens the lanes, and init creates the workspace under .pi/Teamwork/Teams/<Team>/ without touching your tree. When you want to see or steer the engine yourself:

$ teamwork map
$ teamwork criteria add --id AC-1 --text "CLI prints the version and exits 0" --owner developer --by ceo
$ teamwork loop

criteria add records what "done" means; loop is the mechanical stop check and reports CONTINUE until every criterion and test layer passes. Full walkthrough: Quickstart.

Features

  • 15 role charters (CEO, CTO, COO, CFO, Architect, Product Manager, Project Manager, Designer, Developer, Tester, Release Engineer, Technical Writer, Data Engineer, Reviewer, Security Auditor).
  • Role-shared memory and task queues; private per-instance workspaces.
  • Two-stage quality gate: author self-check, then independent Reviewer verdict.
  • Layered tests, one owner per layer, tracked in tests plan.
  • Hash-chained ledger with ledger verify, ledger export and trace.
  • Provider slot leases for concurrency caps.
  • Portable runtime: CPython 3.9-3.13 on Linux, macOS and Windows.

Documentation

Page Read it for
Installation Install the skill into the directory your agent host scans; optional engine package.
Quickstart Bootstrap a team in an empty project and run the CEO loop.
CLI reference Every command, subcommand, flag and exit code.
Architecture Package layout, resource resolution, workspace and ledger.
CI pipeline Run the full lint/test/build/docs pipeline locally.

Repository