菜单
Governed autonomous OODA loop for cognitive agents. Use when starting, stepping, monitoring, or stopping agent sessions with human checkpoints and safety gates.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Compile mode_hooks definitions into the plugin_projector pipeline. Use to inspect mode_hooks status (read-only in stage 0).
Manage Claude Code plugin projection from KairosChain SkillSets. Use when checking projection status, forcing re-projection, or verifying integrity.
Scaffold and design new SkillSets. Use when creating new SkillSet packages with tools, knowledge, and optional Claude Code plugin artifacts.
Exchange SkillSets between KairosChain instances via the decentralized Meeting Place. Use when acquiring, depositing, withdrawing, or browsing SkillSets from peers.
Self-amendment MCP server framework with layered skill architecture and blockchain auditability. Use when managing knowledge layers, evolving skills, auditing health, or understanding KairosChain's architecture and capabilities.
| name | agent |
| description | Governed autonomous OODA loop for cognitive agents. Use when starting, stepping, monitoring, or stopping agent sessions with human checkpoints and safety gates. |
Manage autonomous agent sessions with observe-orient-decide-act-reflect cycles.
agent_start goal="..." — create mandate, run OBSERVEagent_step — run ORIENT → DECIDE → ACTagent_step or agent_stopagent_start goal="..." autonomous=true — start with auto-cyclingagent_status — check progress at any timeagent_stop — terminate when done or if pausedllm_call adapter chain)The agent's OODA phases invoke llm_call (from the llm_client dependency)
which spawns external LLMs as subprocesses via adapter classes:
| Adapter | Subprocess command | Use case |
|---|---|---|
ClaudeCodeAdapter | claude -p --output-format json | Sub-author (4.6), persona reviewers |
CodexAdapter | codex exec --sandbox read-only | Codex review |
CursorAdapter | agent -p | Cursor review |
AnthropicAdapter | Direct API (no subprocess) | Anthropic API calls |
OpenaiAdapter | Direct API | OpenAI API calls |
The agent CAN orchestrate multi-LLM review, invoke sub-author processes, and leverage cross-provider reviewers — all from within the governed OODA loop with blockchain recording of each step.
external_tools SkillSet)SafeFileWrite and SafeFileEdit are available via invoke_tool, enabling
the agent to write design drafts to docs/drafts/ or other project paths.
All KairosChain MCP tools (context_save, multi_llm_review, chain_record,
knowledge_get, etc.) are available via invoke_tool in the Act phase.
/kairos-chain:agent-monitorPost-session review agent. Invokes agent_status, autonomos_status, and chain_history
to summarize session progress and flag anomalies. Read-only — cannot modify state.