overture

Scaffold a new Claude Code skill into this plugin (or the current project). The user names the skill and picks a category — agent review, visualizer, documentation, other — and the matching SKILL.md template lands on disk, ready to edit.

Name this Claude Code session as an owner-console agent, e.g. /overture:as agent-5 (typed by the user only).

Use whenever work produces a question the owner must decide (a study's open questions, a lane brief's owner questions, a review trade-off) in a project registered for the owner console - post each as a lockable question so it reaches the owner's inbox, instead of leaving it in a document.

Use to turn the owner console's locked answers into the project's record - export them to committed files, preview the fold, fold, and land it through a PR. Called by console-process, or on its own.

Use to run one owner-console deliberation (a "fork", its follow-up round, a roar panel on one answer, a refine or drill, or a deliberation on an open question before the owner answers it) - build the round's bundle, sit the committee or run the project's skill, and write back one to five lockable questions plus one summary message (or, before an answer, one recommendation reply). Called by console-process for every waiting fork.

Use when the user wants to set up the owner console for the current project (onboard, install, connect it to Cloudflare), or asks how to get the console running. Asks for the project name and the Cloudflare values, writes the project's non-secret config with the kit's onboard.py, and hands the user the commands only they may run.

Use when the owner console's doorbell says "process" (the owner pressed "Answers are in") or "chat" (a chat message in the inbox) or "visual" (a visual request), when the SessionStart note lists owner requests, or when a background `agent.py watch` exits. Folds newly locked answers through a PR, replies to threads and chat messages awaiting the agent, runs waiting forks (roar, refine and drill included) and visual requests, marks the console synced and re-arms the watch.

Use when the owner console asks for a UX component - a visual request on a UX item (one whose section has a lane-ui or lane-ux segment). Designs and codes the component as a world-class UX designer and component engineer - accessible, psychologically sound, on-trend without being faddish - as ONE self-contained HTML document that works with no JavaScript, checks it with tools/ux_check.py, posts it with `agent.py visual`, and lands it by PR. Called by console-process for UX visual requests instead of console-visual.

Use when the owner console's doorbell carries a `visual` request (the owner pressed "Request a visual" on an item) - draw it as a Mermaid diagram or a static, self-contained HTML mock with a short doc, store it with `agent.py visual`, and land it by PR from your own worktree with `agent.py visual-export`. Called by console-process for every waiting visual request.

Work through one decision with the owner - frame what depends on it, lay out two to four options with their real costs, recommend one with a star, and name what would change the recommendation. Use when the user types /overture:deliberate (with a question, or "open" to list the undecided items), or when an owner decision needs laying out before it is made.

Drill one level down into a segment or concept that has no spec yet - list the functions it must capture, ask the three to five questions that would change its spec (each with lettered options and a star), then write a short standalone spec. Use when the user types /overture:drill, or for an owner-console "Drill" fork (the overture:drill next_step).

Adversarially grill the operator on a half-formed idea BEFORE it spawns work. Produces 3–5 pointed questions that stress the idea's user, failure mode, scale, opportunity cost and evidence. Writes each as a question against the current item so the operator can answer them in-console. Inspired by mattpocock's "grill me" skill, re-shaped for Overture's question/answer/lock pipeline. Use when the owner types /overture:grill-me, or when the Launch Idea pane in the console asks for machine-authored grills.

Use when the user wants to run a named playbook on this project's console — a codified multi-step delegation (round, visual, chat). The playbook files live under .overture/playbooks/<slug>.json. This skill lists what is available, confirms which to run, and dispatches it through the console's agent socket; each step then arrives in the owner's inbox like any other delegation.

Revise one existing spec or planning document so it says what was decided - read it and the decisions and findings that bear on it, propose the changes as a change list before touching it, and name the other documents the change ripples into. Use when the user types /overture:refine, or for an owner-console "Refine" fork (the overture:refine next_step).

Run a three-perspective panel (architect, ux, and an advisor seat) on a question, decision or design - independent reads in parallel, a deliberation round where the panelists challenge each other, then a full implementation plan. Use when the user types /overture:roar followed by a question, decision or design under review. Inside an owner-console round (a fork with roles ["roar"]) the console-fork skill's section 7 governs, and the round ends in lockable questions, not a plan.