docs-harness
name: README description: Use when understanding project overview, directory responsibilities, or basic usage.
docs-harness
English | 中文
docs-harness is an agent-native project documentation management tool. It helps agents reliably discover, read, validate, and continuously maintain project documentation.
Usage
Copy this prompt to your agent:
Install docs-harness first:
npm i -g docs-harness
Then run this in the target project:
docs-harness skills read agent-init
Read the returned content and follow that workflow to initialize docs-harness,
confirm the document management scope, validate the document graph, and repair
any document issues.Project Characteristics
docs-harness is designed for agents, not as a traditional human-facing documentation site.
This means the agent can take over project documentation the same way it takes over project code: creating, reading, updating, moving, and deleting documents as part of normal project work. Instead of asking humans which Markdown file to read, the agent uses the docs-harness document protocol to discover, read, validate, and maintain the right project documents.
During adoption, docs-harness first asks the agent to confirm which Markdown files should enter the managed documentation scope. Documents inside that scope can later be updated, migrated, split, deleted, or indexed by the agent. Documents that are not ready to be managed can be skipped for now and brought under management later.
How It Works
docs-harness organizes project documentation as a document graph.
When an agent discovers that a module is a complete independent function, it can
create a README.md for that function and a route file that distributes the
document index for that function. A module can be a package, module, service,
subsystem, or even the entire project.
The route file depends on the agent type:
- Claude uses
CLAUDE.md - Other agents use
AGENTS.md
Routes from different independent functions connect to each other and form the project documentation graph. When agents discover, read, update, or validate documents, they follow this graph instead of guessing Markdown file paths.
Reading
Agents use insight to understand which documents are relevant to the current
location.
insight returns the current function's README description and the document
index from its route. The agent uses those entries to decide which documents are
relevant to the task, then reads the full document by stable name.
Writing
docs-harness writes documents through document types.
Each document type has a clear role, such as function README, route, runbook, architecture note, or long-lived constraint. Before writing, the agent reads the current project's document type configuration and chooses the type that matches the task.
If the built-in document types do not fit your project, ask the agent to define new types or update existing ones for you.
Automatic Maintenance
During daily document reads and writes, docs-harness collects improvement signals.
These signals represent document friction discovered during real use, such as a
document that is not indexed correctly, a description that no longer matches how
the document is used, or content that is too large for its role. A recurring
maintenance loop can run schedule-document-quality-maintenance to process
these signals and keep documentation quality aligned with the actual project.
CLI Commands
docs-harness is an AI-native CLI; agents know which CLI command to run for each document task, and humans do not need to learn the command surface.