ato

How an AI agent converts mockup HTML into ato-desktop's GPUI Rust components. Use the external `gpui-html` CLI to lower HTML to a Rust builder skeleton; integrate it into production components by hand — do NOT add an HTML parser to ato-desktop.

Deploy and release ato-cli using gh and sleep across dev, main, and tagged GitHub Releases. Use when pushing release-related changes, creating dev-to-main PRs, monitoring required GitHub checks, merging after green, dispatching the Release PR or release-tag workflow via workflow_dispatch, handling release-plz PRs, creating patch bump PRs after version collisions, and verifying the published GitHub Release assets.

Release ato-cli or nacelle end-to-end using gh and sleep until the final GitHub Release is published. Use when pushing release-related changes, creating and monitoring PRs, handling ato-cli release-plz PRs, tagging ato-cli or nacelle, watching release workflows, verifying GitHub Release assets, and recovering from version or tag collisions.

Open a GitHub issue on ato-run/ato distilled from the current session. Use when the user says "issueを作成", "issue化", "open an issue", "file a bug", or otherwise asks to capture a problem/idea uncovered in conversation. Writes the issue in English, attaches the right area:/topic:/platform: labels, and uses the standard ato body template.

Cut a new ato-run/ato release. Merges dev into main, bumps the four dist-tracked crates in lockstep, tags from main, monitors desktop-release.yml + cargo-dist release.yml, and recovers from known failure modes (frontend dist, 403 on host, draft/published race, same-version retry). Use whenever the user says "リリースする", "devの変更をmainに入れてリリース", "cut a release", or `curl install.sh` is broken because of missing assets.

Drive the running ato-desktop (GPUI/Focus mode) end-to-end through the ato-desktop-mcp bridge to exercise real user flows: open a capsule, satisfy config/consent/secret gates, reopen an installed app via ato://app/<ipk>, stop/restart sessions, and inspect both guest WebViews and the GPUI host surface. Use for AODD-style operation, not unit tests.

Agent Operation Driven Development (AODD). Drive product development from agent operation reports instead of pre-written specs or assertions — the operation itself is the development input, the way a failing test is in TDD. Have an autonomous agent operate the real user-facing surface of an application, observe where it gets stuck, and turn those frictions into a Usecase Receipt that becomes the next development input. USE FOR: agent-driven UX evaluation, end-to-end feasibility checks for real users, finding usability and onboarding frictions an agent (or first-time user) actually hits, prioritizing fixes by what blocks autonomous completion. DO NOT USE FOR: unit-test authoring, pure regression coverage of known flows (use E2E tests instead), or one-shot bug fixes with a known repro.

Run hermetic verification for ato-cli and ato-desktop via MCP without polluting the developer's real ~/.ato state. Covers isolated env setup, focused cargo tests, CLI run/session verification, desktop socket discovery checks, and stdio MCP smoke tests.

Manage ato-run/ato work in GitHub Projects using gh project with a minimal operating model. Covers auth, project bootstrap, field setup, issue or PR intake, triage queries, release audits, and the handoff point to built-in workflows or gh api graphql when gh project stops being the right tool.