/install genor-orchestrator
Genor's Project Orchestration
Single source of truth for agentic development work, model orchestration, and project management.
Core Principles
- Codebase First — Never touch code without understanding it first
- Plan Before Act —
update_planfor anything beyond one edit - Verify Before Claim — No completion claim without fresh evidence
- Self-Review — Every output gets audited before delivery
- Fail Gracefully — Every step has a fallback chain
- Document Everything — Log sessions, decisions, architecture
Data Directory
All user data lives in orchestrator-data/ (set ORCHESTRATOR_DATA_DIR to override).
orchestrator-data/
├── models.json — model inventory
├── session_log.md — session history table
├── price_changes.log — price tracking
├── MODEL_CATALOG.md — generated catalog
├── projects/ — per-project session tracking
│ └── \x3Cname>/sessions.json
└── sessions/ — detailed session state files
└── YYYY-MM-DD-HHMM-\x3Cproject>-\x3Ctask-slug>.md
Scripts
| Script | Purpose |
|---|---|
bash {baseDir}/scripts/onboard.sh |
First-time setup |
bash {baseDir}/scripts/init-project.sh \x3Cpath> \x3Cname> [stack] |
Scaffold project |
bash {baseDir}/scripts/log-session.sh ... |
Log session |
bash {baseDir}/scripts/log-decision.sh ... |
Log ADR |
bash {baseDir}/scripts/check-prices.sh |
Price check |
bash {baseDir}/scripts/discover-models.sh |
Probe providers |
bash {baseDir}/scripts/test-model.sh \x3Cid> |
Test connectivity |
bash {baseDir}/dashboard/serve.sh |
Start dashboard |
Conversational Triggers
- "onboard" →
bash scripts/onboard.sh - "start dashboard" →
bash dashboard/serve.sh - "onboard project X" →
bash scripts/init-project.sh \x3Cpath> \x3Cname> - "check prices" →
bash scripts/check-prices.sh
Execution Workflow
Phase 0: Init
Understand the codebase (exec find . -name '*.ts' | head -50), search memory, research.
Phase 1: Plan
Call update_plan. Size the work: small (1-3 files), medium (3-8), large (8+ → decompose).
Phase 2: Execute
| Scenario | Primary | Fallback |
|---|---|---|
| Single-line edit | edit tool |
exec sed |
| Multi-file change | ACP coding agent | Manual edit |
| Research | Sub-agent | Web search |
| Debugging | Phase 5 protocol | Mental trace |
| Testing | Test framework | Manual |
| Browser testing | Browser tool | curl |
Phase 3: Verify
Build → Test → Lint → Screenshot (if UI). No claim without evidence.
Phase 4: Manage
Update .planning/STATE.md and .planning/ROADMAP.md after every session.
Phase 5: Diagnose
Reproduce → Hypothesise (3-5 causes) → Instrument one variable → Fix → Regression test.
Phase 6: Tool Fallbacks
| Tool | Fallback chain |
|---|---|
| Codebase discovery | exec find → exec ls -R → read key files |
update_plan |
mental plan → STATE.md note |
| ACP coding agent | CLI variant → sub-agent → manual edit |
edit tool |
exec sed → write full file |
| Build | tsc --noEmit → node --check |
| Test | specific test file → manual |
| Vision | cloud → local → describe |
| Memory search | lcm_grep → exec grep |
Model Routing
| Task Type | Primary | Fallback 1 | Fallback 2 |
|---|---|---|---|
| Heavy coding | Best available | ACP agent | Fast cloud |
| Quick edits | Fast cloud | Free tier | Local |
| Research | Best reasoning | Free tier | Local |
| Vision | Cloud vision | Local | Describe |
| Planning / design | Best reasoning | Free tier | Fast cloud |
| Docs / summaries | Free tier | Fast cloud | Local |
See {baseDir}/ROUTING.md for the current routing table. See orchestrator-data/MODEL_CATALOG.md for full inventory.
Sub-Agent Protocol
Every spawned sub-agent MUST include:
IMPORTANT: Follow orchestration conventions:
- Understand the codebase first (exec find as fallback)
- Plan before coding (update_plan or mental plan)
- Verify before claiming (build, test, screenshot)
- Self-review output before returning
- Use fallback chains when tools are unavailable
Design Grilling
Before significant architectural work, conduct a structured interview:
- One question at a time, wait for feedback
- Challenge terms that conflict with existing CONTEXT.md
- Propose precise canonical terms for vague language
- Stress-test with edge cases
- Cross-reference with code; surface contradictions
Generate: CONTEXT.md (glossary), .planning/ADRs/ (decisions), summary.
ADR criteria (all three must be true): hard to reverse, surprising without context, real trade-off.
References
| Resource | Path |
|---|---|
| Full documentation | {baseDir}/references/README.md |
| Onboarding guide | {baseDir}/references/ONBOARDING.md |
| Execution reference | {baseDir}/references/EXECUTION.md |
| Debugging guide | {baseDir}/references/DEBUGGING.md |
| Fallback tables | {baseDir}/references/FALLBACKS.md |
| Routing table | {baseDir}/ROUTING.md |
| Model catalog | orchestrator-data/MODEL_CATALOG.md |
| Dashboard | http://localhost:8766 (when running) |
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install genor-orchestrator - 安装完成后,直接呼叫该 Skill 的名称或使用
/genor-orchestrator触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
Genor's Project Orchestration 是什么?
Unified orchestration: model routing, coding workflow, planning, verification, and project management. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 37 次。
如何安装 Genor's Project Orchestration?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install genor-orchestrator」即可一键安装,无需额外配置。
Genor's Project Orchestration 是免费的吗?
是的,Genor's Project Orchestration 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
Genor's Project Orchestration 支持哪些平台?
Genor's Project Orchestration 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。
谁开发了 Genor's Project Orchestration?
由 Krzysztof(@genortg)开发并维护,当前版本 v2.2.1。