openclaw-clarke

CLARKE (Cognitive Learning Augmentation Retrieval Knowledge Engine) plugin for OpenClaw.

Gives your OpenClaw agent persistent memory, learned context, and self-improving skills — all injected programmatically into every LLM call.

Install

openclaw plugins install @claw_clarke/openclaw-plugin

What It Does

• Every LLM call gets CLARKE context injected into the system prompt (policies, decisions, skills, domain knowledge)
• Every user query gets retrieval-augmented with relevant documents, memory, and decisions
• Every response feeds back into CLARKE's learning loop for continuous improvement

Hooks

| Hook | Purpose | |------|---------| | agent:bootstrap | Inject CLARKE identity into bootstrap files | | before_prompt_build | Inject session context (cached 60s) into system prompt | | before_agent_reply | Query-specific RAG augmentation | | session_start | Greeting + cache invalidation + auto-registration | | llm_output | Implicit feedback for learning loop |

Tools

| Tool | Description | |------|-------------| | clarke_status | System health, agents, policies | | clarke_recall | Query CLARKE memory | | clarke_teach | Submit knowledge and corrections | | clarke_review | List pending directive proposals |

Commands

| Command | Description | |---------|-------------| | /clarke | Dashboard | | /clarke_recall <question> | Query memory | | /clarke_teach <knowledge> | Teach CLARKE |

Prerequisites

CLARKE backend must be running. See the setup guide.

Configuration

Set these environment variables (or configure via openclaw.json):

CLARKE_API_URL=http://localhost:8000
CLARKE_TENANT_ID=<auto-registered>
CLARKE_PROJECT_ID=<auto-registered>
CLARKE_AGENT_SLUG=clarke-operator

Tenant and project IDs are auto-registered from the workspace path on first session start.

License

Polyform Noncommercial 1.0.0 — free for personal and noncommercial use. For commercial licensing: [email protected]