エンジニアリング
Personal software factory: Grok Bot orchestrates, Jev scores the loops
Cloud coding agents ship slop when nobody gathers context first or runs real review loops.
困っていたこと
Ian Lapham (@ianlapham) uses Grok Bot as the orchestrator for a personal software factory: it ingests Slack, GitHub, and Notion, then kicks cloud coding agents through a graph of loops with Jev as verifier at each step. Pipeline: (1) Orchestration — PM + engineering agents gather live context and start work; (2) Planning/implementation on strong cloud models; (3) Review pass — parallel context-free sub-agents (minimize, repo prefs, bugs, anti-slop, evidence) report to Jev; critical findings go back to a coder; (4) Review pass 2 with a different model family; (5) Optional on-device testing on a remote Mac mini (pre-warmed devices, auth skills, feature-vs-plan check); (6) Fresh intent check with Jev on screenshots/video; (7) PR skill uploads evidence, watches CI bug bots, and re-loops on failure. Tip with author-published steps. Not affiliated with xAI.
インストールまたは作成するもの
Grok Bot context (Slack/GH/Notion) → cloud plan/code → multi-loop Jev review → optional device test → intent check → PR + CI loop.
ボット
PM orchestrator
Gather Slack/GitHub/Notion context; kick planning when you are ready to start work.
Engineering orchestrator
Hand context to cloud coding agents and drive the review/test/ship loops.
Jev verifier
Score review findings and intent checks; send critical misses back to a coder.
プラグイン
Notion
必須
GitHub
必須
ファイル
Planning skills (full-stack vs app-specific env setup); PR creation skill that attaches on-device evidence.
ルーティン
初回の実行では使いません。まず手動で1回通して確認してください。
プロンプト
これはヒントであり、貼り付けられるシステムプロンプトではありません。作者はプロンプトを公開していません。初回の手順に従ってください。
初回実行
- 1Read the thread starting at https://x.com/ianlapham/status/2104690437650857993.
- 2Pipe Slack, GitHub, and Notion into Grok Bot with skills that keep context live; run separate PM and engineering kickoff jobs.
- 3Have cloud agents plan and implement with strong models, then run a context-free parallel review pass (minimize, prefs, bugs, anti-slop, evidence) scored by Jev.
- 4Run a second review pass on a different model family; optionally test on a pre-warmed remote Mac with auth/nav skills.
- 5Finish with a fresh intent check (screenshots/video + Jev), then open a PR with evidence and re-loop until CI/bug-bots pass — you approve merge.
承認フェンス
Merge and production ship stay human-gated; agents may open PRs and re-loop CI only inside your rules.
調整ポイント
Steal “context first, then parallel context-free review loops scored by a verifier.” Humans step in at the end, not every commit.
出典
Ian Lapham @ianlapham · 元の投稿
xAI とは提携していません。このプロンプトは当サイトが書いたものではありません。第三者テンプレートの規約も適用されます。
Pro パック
アイデアだけでなく、完全な構成が必要ですか？
Pro パックには、人気ユースケースのジョブ本文、プロンプト、スケジュール、落とし穴が入っています。