motocortex

all skill you need for agent coding.

motocortex is TokenRoll's general-purpose skill/plugin for development. It is not tied to any one product — drop it into any repo to turn a rough idea into a set of documents an autonomous agent can run on its own.

Read in another language: 简体中文

What it does

You run one command, /start, and it walks you through the setup: it reads the current state of your project, interviews you about the decisions you may not realize are still open, and lands five documents — VISION.md / DOR.md / DOD.md / LOOP.md / PROGRESS.md. After that, /goal drives the repo round after round: an agent picks an unchecked item, verifies it with reproducible evidence, checks it off, and keeps going until everything is done or only blockers remain.

Flow

/start                                   ← authoring phase: the one command to remember
  1. gather context   (read the repo, existing conventions, the settled stack)
  2. /grill           ← interview until nothing is silently assumed → land DECISIONS.md
  3. /frame           ← produce VISION.md (what to build) + DOR.md (readiness risk list)
  4. /loop            ← from the goal, generate DOD.md (what counts as done)
                        + LOOP.md (how each round runs) + PROGRESS.md (the ledger)

/goal                                    ← execution phase: drive the repo autonomously
  run the LOOP state machine round by round: pick a DoD item → verify with
  reproducible evidence → check it off, until everything's done or only blockers
  remain (not installed as a command? paste the "How to start" block at the top
  of LOOP.md into any agent session)

Skills

skill	invocation	responsibility	artifacts
start	user	orchestration entry: chains the steps below	all five documents
grill	model	gather context + interview for the unasked questions until the frontier is empty	DECISIONS.md
frame	model	synthesize DECISIONS into the source of truth and the readiness list	VISION.md DOR.md
loop	model	generate the autonomous execution engine from the goal	DOD.md LOOP.md PROGRESS.md
goal	user	drive a prepared repo through the LOOP state machine to done or blockers	appends PROGRESS.md

Two layers: start/goal handle orchestration (user-invoked), while grill/frame/loop are reusable discipline (pulled in by start).

Design principles

• Look up facts, leave decisions to the user. Anything discoverable from the environment gets looked up, not asked. Real judgment calls go to you — one at a time, each with a recommended answer.
• Don't re-litigate what's settled. An existing mature stack or convention is recorded and confirmed, not re-interviewed.
• Checking off means evidence. Every DoD item is checked off on reproducible evidence — a command and its output, not "I think it's done."

Install

As a Claude plugin (v0.0.1):

/plugin marketplace add TokenRollAI/motocortex
/plugin install motocortex

Or symlink the directories under skills/ into your skills directory (e.g. ~/.claude/skills).