The Missing Intelligence Playbook

Intelligence over vibes. Your OpenClaw agent has personality dialed in but the operations manual missing.

Your agent uses the same model as Claude Code. So why does Claude Code feel smarter?

It's not the model. It's the harness. Claude Code's system prompt contains ~5,000 words of prescriptive operational instructions — "use tool X for task Y", "spawn sub-agents when Z", "never brute-force a failing approach." Most OpenClaw agents have personality dialed in but the operations manual missing.

This package adds that operations manual.

What's Inside

AGENTS.md operational sections (inherited by sub-agents):

• Tool routing rules — stops your agent from shelling out cat when it has a file read tool
• Sub-agent delegation framework — when to spawn, model selection, task specification template
• 4-pass research methodology — broad scan, deep dive, skeptical cross-reference, actionable synthesis
• Self-configuration rules — read before edit, check docs, verify keys exist
• Quality gates — checklists for research, code, and all output
• Output discipline — front-load answers, confirm completion

TOOLS.md complete reference:

• Full tool inventory with when-to-use guidance
• Tool chaining patterns (research-to-delivery, code-test-commit, config-verify)
• Known gotchas (gateway restarts, memory indexing lag, sub-agent inheritance)
• Search strategies for code, past context, and documentation

SOUL.md additions:

• Multi-pass research methodology
• Insight development process
• Decision-making heuristics

openclaw.json config changes:

• Main timeout: 600s to 1800s
• Sub-agent timeout: 3600s
• Spawn depth: 2 (agents can spawn agents that spawn agents)
• Max children: 8
• Loop detection enabled

The Key Insight

"Be resourceful" is aspirational. "Use web_search for internet queries, web_fetch for reading URLs, browser only for interactive pages" is operational. AI agents need both. This package adds the operational half.

Who It's For

• OpenClaw users whose agent feels "dumber" than Claude Code despite the same model
• Anyone whose agent shells out cat and grep instead of using built-in tools
• Anyone whose agent does everything single-threaded instead of spawning sub-agents
• Power users who want their agent to match Claude Code's operational intelligence

Installation: Under 10 minutes. Two paths: fresh workspace (copy files directly) or existing workspace (merge operational sections). Config changes require a gateway restart.