Guide: the slash commands (bring your own assistant)
ASDD has three ways in: the CI gates, the CLI, and the Goose operate kit. The slash commands are the
CLI's front door inside a coding assistant. They ship for Claude Code under
.claude/commands/asdd/, but the substance is in the CLI they call, not
in the prompts, so they port to any assistant in minutes.
That is the whole point: ASDD does not live inside your assistant. The commands are thin wrappers
over asdd ...; the gates, the config, and the pipeline are files in your repo. Bring Claude Code,
Cursor, Copilot, or your own harness - the governance is identical.
The four commands
| Command | What it does | The CLI underneath |
|---|---|---|
/asdd:setup | Walks a maintainer through wiring a model to each agent role, in plain language, and refuses a developer == tester assignment. | asdd setup |
/asdd:spec | Turns an idea into a spec that passes the intake gate. It fills the definition of ready (outcomes, scope, constraints, verification), writes it where spec_paths: says, and checks it with the same gate the pipeline runs. On an OpenSpec project it defers to openspec. | asdd spec-check (or openspec validate) |
/asdd:review | Runs the review the way the pipeline will, before you push: the deterministic gates first, then the lenses if a model is wired. | asdd openspec-gate / the review lenses |
/asdd:status | Shows the governance state: what the agents are doing and under what steering, from the read-only dashboard. | asdd dashboard |
Why a spec that passes locally passes on the PR
/asdd:spec and /asdd:review call the same deterministic gates the CI pipeline calls - there is no
second implementation to drift. A spec that asdd spec-check accepts on your machine is accepted at
intake; a change the local review flags is the change the pipeline flags. The commands are a faster
feedback loop over the identical code, not a preview of it.
Porting to another assistant
Each command is a short Markdown prompt whose only job is to invoke a CLI subcommand and read back its output. To bring your assistant of choice:
- Install the CLI:
pip install git+https://github.com/OneHillAI/ASDD(orpip install .from a checkout; the wheel is self-contained). - Copy the four prompts from
.claude/commands/asdd/into your assistant's command format. The bodies are plain instructions; only the file location and front matter are Claude-Code-specific. - The commands assume
asddis onPATH. That is the only dependency.
Steering: read on the dashboard, write via the CLI and config
The commands are for doing (draft a spec, run a review, wire a model). To steer - change what the
gates require, which paths are protected, which model a role runs on - edit .asdd.yml and the
recipes/, and read the current state with /asdd:status. Nothing steers by chat: the config is the
source of truth, and the agents pick it up from there. See
gates-and-requirements.md for every knob.