2f57ef8948
- AgentHub: 13 files updated with non-engineering examples (content drafts, research, strategy) — engineering stays primary, cross-domain secondary - AgentHub: 7 slash commands, 5 Python scripts, 3 references, 1 agent, dry_run.py validation (57 checks) - Marketplace: agenthub entry added with cross-domain keywords, engineering POWERFUL updated (25→30), product (12→13), counts synced across all configs - SEO: generate-docs.py now produces keyword-rich <title> tags and meta descriptions using SKILL.md frontmatter — "Claude Code Skills" in site_name propagates to all 276 HTML pages - SEO: per-domain title suffixes (Agent Skill for Codex & OpenClaw, etc.), slug-as-title cleanup, domain label stripping from titles - Broken links: 141→0 warnings — new rewrite_skill_internal_links() converts references/, scripts/, assets/ links to GitHub source URLs; skills/index.md phantom slugs fixed (6 marketing, 7 RA/QM) - Counts synced: 204 skills, 266 tools, 382 refs, 16 agents, 17 commands, 21 plugins — consistent across CLAUDE.md, README.md, docs/index.md, marketplace.json, getting-started.md, mkdocs.yml - Platform sync: Codex 163 skills, Gemini 246 items, OpenClaw compatible Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
3.7 KiB
title, description
| title | description |
|---|---|
| Agent Orchestration — Multi-Skill Coordination Protocol | A lightweight protocol for orchestrating AI coding agents, personas, and skills across domains. Coordinate Claude Code skills and Codex agents on complex, multi-domain work. |
:material-sitemap: Orchestration
A lightweight protocol for coordinating personas, skills, and agents on work that crosses domain boundaries. No framework required. No dependencies. Just structured prompting.
Core Concept
Most real work crosses domain boundaries. A product launch needs engineering, marketing, and strategy. An architecture review needs security, cost analysis, and team assessment.
Orchestration connects the right expertise to each phase of work:
- Personas define who is thinking (identity, judgment, communication style)
- Skills define how to execute (steps, scripts, templates, references)
- Task agents define what to do (scoped, single-domain execution)
Patterns
Solo Sprint
One person, one objective, multiple domains. Switch personas as you move through phases.
Week 1: startup-cto + engineering skills → Build
Week 2: growth-marketer + marketing skills → Prepare launch
Week 3: solo-founder + business skills → Ship and iterate
Best for: side projects, MVPs, solo founders.
Domain Deep-Dive
One domain, maximum depth. Single persona, multiple skills stacked.
Persona: startup-cto
Skills loaded simultaneously:
- aws-solution-architect (infrastructure)
- senior-security (hardening)
- cto-advisor (tech debt assessment)
Task: Full technical audit of existing system
Best for: architecture reviews, compliance audits, technical due diligence.
Multi-Agent Handoff
Different personas review each other's work.
Step 1: startup-cto designs the architecture
Step 2: growth-marketer reviews from user/market perspective
Step 3: solo-founder makes the final trade-off decision
Best for: high-stakes decisions, launch readiness reviews, investor prep.
Skill Chain
No persona needed. Chain skills sequentially for procedural work.
1. content-strategy → Identify topics and angles
2. copywriting → Write the content
3. seo-audit → Optimize for search
4. analytics-tracking → Set up measurement
Best for: repeatable processes, content pipelines, compliance checklists.
Example: 6-Week Product Launch
| Phase | Weeks | Persona | Skills | Output |
|---|---|---|---|---|
| Build | 1-2 | startup-cto | aws-solution-architect, senior-frontend | Architecture doc, deployed MVP |
| Prepare | 3-4 | growth-marketer | launch-strategy, copywriting, seo-audit | Landing page, content calendar |
| Ship | 5 | solo-founder | email-sequence, analytics-tracking | Launch, tracking verified |
| Iterate | 6 | solo-founder | form-cro, copy-editing | Conversion improvements, metrics report |
Rules
- One persona at a time. Switching is fine, but don't blend two in the same prompt.
- Skills stack freely. Load as many as the task needs.
- Personas are optional. For procedural work, skill chains alone are sufficient.
- Context carries forward. When switching phases, summarize decisions and artifacts.
- The human decides. Override any phase, persona, or skill choice.
Quick Reference
Activate a persona:
Load agents/personas/startup-cto.md
Load a skill:
Load engineering/aws-solution-architect/SKILL.md
Phase handoff:
Phase 1 complete.
Decisions: [list]
Artifacts: [list]
Open items: [list]
Switching to: [persona] + [skills]