Agent Workflow

AI coding assistants work with dot•requirements through the CLI — searching, validating, composing, reviewing, and tracking coverage without leaving your development environment. There’s no server to install: dotreq ai-setup writes workflow guidance into your assistant’s context file, and the guidance tells the agent which commands to run and when.

Setup

Run the interactive setup:


dotreq ai-setup

This will:

Prompt you to select your AI assistant
Install the requirements-driven workflow guidance into your project’s context file
Remove any MCP configuration written by earlier versions (the local MCP server is retired)

Supported Assistants

Assistant	Context File
Claude Code	`CLAUDE.md`
Cursor	`AGENTS.md`
OpenAI Codex	`AGENTS.md`
Google Antigravity	`AGENTS.md`
GitHub Copilot	`AGENTS.md`

The section is wrapped in HTML comment markers () for idempotent updates — running ai-setup again updates the section without affecting the rest of your context file.

The Workflow Guidance

The installed guidance is always-on: it’s in every conversation’s context, so the agent doesn’t have to remember to look anything up. It covers:

The workflow — the 8-step loop from finding requirements through test review
Requirements syntax — enough to write valid .requirements.md files
Test usage — how to reference requirements with requirement()
CLI verbs — which commands to use for exploration, authoring, review, and coverage

How Quality Review Works

The two judgment steps — style-reviewing requirements and reviewing tests against requirements — run as subagent dispatches:

The agent dispatches a subagent (fresh context, no attachment to the draft)
The subagent runs the review verb, which emits everything a reviewer needs:


dotreq style-check .requirements/auth.requirements.md
# → the project's style guide + the requirements under review + judgment instructions
 
dotreq review-test src/auth.test.ts
# → the referenced requirement trees + the test content + judgment instructions

The subagent judges the content against the emitted materials and reports findings as MUST FIX / SHOULD FIX / COULD IMPROVE

This runs on your existing AI subscription — no dot•requirements account required. Assistants without subagent support (or CI pipelines) can use the hosted reviewer instead:


dotreq style-check <file> --source cloud
dotreq review-test <file> --source cloud

The hosted reviewer requires a linked project (dotreq link) and applies any custom style guidance configured for your project.

CLI Verbs Reference

Exploration

Verb	Description
`dotreq list [--untested]`	Summary of all requirements (`--untested` filters to coverage gaps)
`dotreq get <id>`	Get a requirement tree with test coverage info
`dotreq search <query> [--regex]`	Search by text or regex across IDs, content, and labels
`dotreq requirements-for <test-file>`	See which requirements a test file references
`dotreq tests-for <req-file>`	Find tests that reference a requirements file

Authoring

Verb	Description
`dotreq create-requirement-document [path]`	Print the format template and your project’s style guide
`dotreq validate [glob]`	Validate `.requirements.md` syntax (works offline)
`dotreq diff [scope...]`	Show how the repo and the cloud differ (read-only)
`dotreq sync [scope...]`	Reconcile the repo and the cloud (both contribute by default)

Review

Verb	Description
`dotreq style-check <file> [--keys ...]`	Emit review materials for a style judgment (`--source cloud` for the hosted reviewer)
`dotreq review-test <test-file>`	Emit review materials for a test-vs-requirements judgment (`--source cloud` for the hosted reviewer)

Coverage

Verb	Description
`dotreq report [--source local\|cloud]`	Coverage report from the local cache or the cloud

CI/CD

Cloud commands in CI authenticate from environment variables automatically when no settings file is present — no flag required:


dotreq sync --repo-contributes

See CI/CD Integration for complete examples.

Working From Chat Instead

Coding agents aren’t the only AI surface: PMs and teammates can drive dot•requirements from Claude and other chat apps through the MCP server — browsing projects, reading specs and coverage, and composing new specs that publish straight to the web app.