メニュー
First-run workspace setup. Use when the user says "set up this workbench", "run setup", or asks how to get started in a fresh clone, to pick runtimes (Claude Code, Codex, or both), personalize local settings, and walk the connector checklist.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/JDWorkdog/ai-workbench --skill setupコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
SOC 職業分類に基づく
| name | setup |
| description | First-run workspace setup. Use when the user says "set up this workbench", "run setup", or asks how to get started in a fresh clone, to pick runtimes (Claude Code, Codex, or both), personalize local settings, and walk the connector checklist. |
Tailor a fresh clone of the AI Workbench to its owner. One repo serves Claude Code users, Codex users, and people who run both; this skill configures (or prunes) accordingly so nobody carries files for a tool they do not use. Everything destructive is approval-gated: propose, list exactly what would change, apply only what the user approves.
personal/docs/setup-receipt.md exists, this is a re-run: read it, say what was configured before, and ask what to change..claude/settings.local.json, .mcp.json, existing personal/ content. Never treat a workspace with journal or draft content as fresh; skip anything already done.Ask once: Claude Code only, Codex only, or both?
Then tailor:
.claude/skills/ is canonical, .agents/skills/ is the mirror, and ./scripts/sync-codex-skills.sh --write runs after any skill edit..codex/ and .agents/. Also update AGENTS.md: drop the sync-script line from AI Harness Checks and the Codex tier column from the routing table. AGENTS.md itself stays; it is the canonical guide that CLAUDE.md imports.AGENTS.md and .agents/skills/ are what Codex reads, so the workspace works as-is. Offer (do not push) a prune of the Claude-specific surface: , , , , and . If pruning, first make canonical: copy over it, then retire the sync script and update to match. Note honestly that the slash-command workflows in have no direct Codex equivalent; the durable capabilities live in skills..claude/commands/.claude/agents/.claude/hooks/.claude/settings*.jsonCLAUDE.md.agents/skills/.claude/skills/AGENTS.md.claude/commands/List every path before removing anything, get an explicit yes, and use git rm so the change is reviewable and reversible.
.claude/settings.local.example.json to .claude/settings.local.json (gitignored) and set USER_NAME. Walk through the permission allowlist and prune it to what the user actually wants..mcp.json.example to .mcp.json (gitignored) and fill in theirs..codex/config.toml for the tier profiles (--profile luna|terra|sol) and note that trust and personal defaults live in ~/.codex/config.toml, not in the repo.The journal pipeline and several commands read external sources. Ask which of these the user has, and have them connect the ones they want in their assistant's connector settings:
Missing connectors are fine; everything degrades gracefully. Record what was connected in the receipt.
Ask whether they want the scheduled journal (the workbench's flagship habit). If yes, walk them through guides/journaling-guide.md: two scheduled tasks in Claude Code Desktop, prompts pasted from the guide, Sonnet-tier model. If no, note that /daily and auto-journaling still work manually and the guide is there when they want it.
python3 scripts/check-ai-harness-rules.py. If Codex is in play and skills were touched, run ./scripts/sync-codex-skills.sh --check.personal/docs/setup-receipt.md: date, runtime choice, files removed (if any), settings personalized, connectors configured, journaling decision. This file is the re-run anchor for Step 1.personal/ (gitignored), and AGENTS.md is where the workspace rules live.personal/ or temp-files/.~/.claude/ or ~/.codex/ user-level config beyond what the user explicitly asks for.Install or upgrade a model-routing and delegation layer in the current project. Use when the user says "tune my harness", "add model routing", "set up delegation", or after a harness cleanup, to add pinned-model agent rosters, a routing table, and QA policy to an existing Claude Code or Codex project.
Answer history questions from the journal system ("when did we meet X", "what did we promise Y", "what happened with Z last fall"). Searches dossiers, then monthly rollups, then targeted dailies, with bounded context at every step. Use for any question about past meetings, commitments, people, accounts, or decisions.