build-mediator

name	build-mediator
description	Orchestrate build steps. Nemotron assesses, dispatches experts, narrates to human, tracks in ops.db.
tags	["build","orchestrate","workshop","mediator"]
version	1.0.0

/build-mediator — Orchestrate a Project Build

When to use

• When a Workshop idea reaches Build stage
• Triggered by tap-daemon.py on "Start Build" button

Input

• idea-id, chat-id, topic-id

Output

• Built project with test link (if deployable)
• ops.db tasks for each step (backbone visibility)
• Telegram narration at each step

Execution

python3 /root/.openclaw/scripts/build-engine.py --idea-id {id} --chat-id {chat} --topic-id {topic}

Steps

1. Assess (Nemotron, <256 tokens) — what type of project?
2. Architect (Claude/Codex) — read project.md, write architecture.md
3. Build (Codex --full-auto) — read architecture.md, create all files
4. Review (Codex) — flag critical issues
5. Deploy (local script) — detect type, serve, return URL
6. Summarize (Nemotron, <256 tokens) — 3 sentences for the human

Rules

• Each step is an ops.db task (agent: "workshop-build")
• Narrate to Telegram BEFORE each step starts (say what you're doing before you do it)
• Nemotron outputs under 256 tokens during orchestration — he decides, doesn't compose
• Agents read files directly — don't stuff context into prompts
• Codex build uses npx directly, NOT codex-task wrapper
• Dynamic config: VPS_IP, ports, model names from env, not hardcoded

Files

• Script: /root/.openclaw/scripts/build-engine.py
• Deploy: /root/.openclaw/scripts/deploy-project.py
• Skills: /root/.openclaw/workspace/skills/build-*/SKILL.md