mupt-ai/dari-docs

Source: https://github.com/mupt-ai/dari-docs

dari-docs

Make your docs so good even the dumbest agent can ship.

dari-docs is a CLI for testing whether your documentation is clear enough for agents to use. It sends your docs to simulated developer agents, asks them to complete real tasks, reports where they get stuck, and can generate proposed docs edits from that feedback.

Use it to turn documentation quality from “seems understandable” into “an agent can actually complete the task.”

Why dari-docs?

Good docs used to mean “a developer can eventually figure this out.” That is no longer enough.

When the reader is an agent, ambiguity becomes measurable. Inconsistent terminology, hidden assumptions, scattered context, and missing setup steps all increase the chance that the agent fails the task or wastes context trying to infer what the docs meant.

dari-docs gives you a repeatable feedback loop for agent-readable documentation: define the task, run simulated users, inspect failures, and optionally pull back edited docs.

What it does

• Tests docs with simulated developers — agents attempt concrete tasks using only the docs you provide.
• Finds task-blocking ambiguity — reports missing context, unclear setup, inconsistent terms, and places where the agent had to guess.
• Generates proposed fixes — optimize turns tester feedback into edited documentation you can review locally.
• Runs managed or self-managed — use the hosted dari.dev Docs service, or run against agents in your own dari.dev org.
• Uses normal agent projects — the tester and editor are just folders of prompts, skills, setup scripts, and a dari.yml manifest.

Install

Install the latest dari-docs binary with the install script, then verify the CLI is available:

curl -fsSL https://raw.githubusercontent.com/mupt-ai/dari-docs/main/install.sh | bash
dari-docs --help

Quickstart

Managed mode uses the hosted dari.dev Docs service and a separate dari.dev Docs credit balance. New accounts start with five dollars worth of free credits.

From your docs repo:

dari-docs auth login

Run a docs check:

dari-docs check . \
  --managed \
  --task "Install the SDK and make a first API call"

The command submits a managed run and prints the run ID. To wait for completion in the same command, add --wait.

Generate proposed docs edits:

dari-docs optimize . \
  --managed \
  --wait \
  --task "Install the SDK and make a first API call"

With --wait, edited files are downloaded into .dari-docs/updated/ without changing your repo. Review that folder and copy changes into your repo when ready.

How it works

1. You point dari-docs at a docs directory or public docs URL and give it one or more tasks.
2. The CLI bundles your local docs, or passes public docs URLs for checker agents to inspect with internet access, then submits a run to hosted Dari Docs agents.
3. Tester agents try to complete the task and report where the docs blocked progress.
4. Use dari-docs runs wait, dari-docs runs download, or pass --wait to collect local run artifacts.
5. If you run optimize, an editor agent proposes documentation changes.
6. Proposed edits can be downloaded to .dari-docs/updated/ for review.

The simulated users are plain dari.dev agents. Managed mode uses the hosted Dari Docs tester and editor agents automatically. If you want to customize the agent prompts, skills, setup scripts, or dari.yml, use self-managed mode.

Managed vs self-managed

Most users should start with managed mode.

Documentation

• Managed mode and billing
• GitHub Actions
• Task files and repeated checks
• Bundle selection
• Live verification secrets
• Agent customization
• Self-managed usage
• Local development