acos-meta
ACOS self-description and configuration skill. Documents how ACOS works, how to extend it, how to add new skills/commands/agents, and how to debug the hook system. Use when building new ACOS capabilities, understanding the system architecture, or onboarding to ACOS for the first time.
frankxai · 已有 AI 审核记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add frankxai/agentic-creator-os --skill acos-meta技能说明
name: acos-meta description: "ACOS self-description and configuration skill. Documents how ACOS works, how to extend it, how to add new skills/commands/agents, and how to debug the hook system. Use when building new ACOS capabilities, understanding the system architecture, or onboarding to ACOS for the first time."
ACOS Meta
ACOS describes itself using the same primitives it uses to build everything else. This skill is ACOS about ACOS.
System Architecture
acos-intelligence-system/
├── .claude-plugin/
│ └── plugin.json ← Plugin manifest (v11.0.0)
├── .mcp.json ← MCP server registry
├── CONNECTORS.md ← Connector category map
├── skills/ ← Domain expertise (subdirs, progressive disclosure)
│ ├── [skill-name]/
│ │ ├── SKILL.md ← Lean main file (<3K words)
│ │ └── references/ ← Deep content, fetched on demand
├── commands/ ← Slash commands (one .md per command)
├── hooks/ ← Lifecycle automation (SessionStart, Stop, etc.)
├── docs/ ← Strategy documents
└── README.md ← Entry point
The 3 Design Principles (Drawn from knowledge-work-plugins)
1. Progressive Disclosure
SKILL.md contains the mental model and workflow skeleton. Details live in references/. Claude loads the lean summary and fetches references only when needed. This keeps context efficient without sacrificing depth.
2. Connector Agnosticism
Skills reference ~~categories, not vendor names. The .mcp.json maps categories to specific tools. Swap tools without touching skill content.
3. Commands as Workflows
Commands are fully-specified workflows in markdown — trigger, input gathering, decision logic, output structure, follow-up options. No code. Claude interprets and executes.
How ACOS Auto-Routing Works
The /acos command is the entry point. It routes based on keyword detection:
User request
|
├── AI architecture keywords → Technical Architect agent
├── Content/writing keywords → Content Engine
├── Music keywords → Music Producer
├── Visual/image keywords → Visual Creation Council
├── Deploy/build keywords → DevOps Pipeline
├── Research keywords → Deep Research swarm
└── Complex/multi-file → Full swarm (5+ agents)
The hook system (hooks/skill-activation-prompt.sh) enhances routing with pattern matching before Claude processes the request.
Adding a New Skill
- Create
skills/[skill-name]/SKILL.md - Add YAML frontmatter:
name,description(include trigger phrases) - Write lean main content (<3K words) covering: overview, core concepts, workflow, key principles
- Create
skills/[skill-name]/references/for detailed content - Register in
skills/skill-rules.jsonif using activation matching - Test: mention trigger phrases and verify activation in session
SKILL.md frontmatter template:
---
name: skill-name
description: "One-sentence description. Include trigger phrases like: what actions activate this skill, what topics it covers."
---
Adding a New Command
- Create
commands/[command-name].md - Use the standard command structure:
- YAML frontmatter:
description,argument-hint(optional) > See CONNECTORS.mdreference## Workflowwith numbered steps- Input gathering, tool use, output format, follow-up options
- YAML frontmatter:
- No code logic — pure markdown workflow
- Test: run
/[command-name]and verify execution
Command frontmatter template:
---
description: What this command does in one sentence
argument-hint: "<optional argument description>"
---
Adding a New Agent
- Create
.claude/agents/[agent-name].md - Define: role, capabilities, tools, escalation path
- Reference from orchestration commands or swarm topology
- Test via Task tool:
Task(subagent_type="[agent-name]", prompt="...")
Debugging the Hook System
ACOS has 15 hooks across 6 lifecycle events. When hooks behave unexpectedly:
- Check audit trail:
cat .claude-flow/audit.jsonl | tail -20 - Check circuit breaker state:
cat .claude-flow/circuit-breaker.json - View learning metrics:
cat .claude-flow/metrics/learning-status.json - Run monitor:
npm run monitor(real-time hook dashboard)
Hook event map:
SessionStart → session-start.js + starlight-bridge + todo-continuation restore
UserPromptSubmit → skill-activation-prompt.sh
PreToolUse → quality-gate + circuit-breaker
PostToolUse → post-tool-track.js + audit-trail
Stop → stop-finalize.js + todo-continuation save + learning-hooks
PreCompact → context preservation
Intelligence Score System
ACOS tracks its own intelligence score across sessions:
| Component | Weight | Measured by |
|---|---|---|
| Skill activation accuracy | 25% | Trajectory success rates |
| Pattern extraction quality | 25% | n-gram count in patterns.json |
| Memory utilization | 20% | Context recovery on session start |
| Hook reliability | 15% | Zero circuit breaker breaks |
| Self-modify safety | 15% | Score delta tracking |
View with /acos-score.
ACOS × knowledge-work-plugins
ACOS v11 integrates patterns from the knowledge-work-plugins ecosystem:
| Pattern | Source | Applied in ACOS |
|---|---|---|
| Progressive disclosure | knowledge-work-plugins | All new skills use SKILL.md + references/ |
| Plugin manifest | knowledge-work-plugins | .claude-plugin/plugin.json |
| Connector agnosticism | knowledge-work-plugins | CONNECTORS.md with ~~category placeholders |
| Command workflow format | knowledge-work-plugins | Standardized command structure with input gathering |
| Two-tier memory | productivity plugin | creator-productivity skill |
| Brand voice framework | marketing plugin | brand |