OpenClaw Deep Research

Run research as a file-backed production workflow, not as a chat-only exercise. Optimize for source traceability, narrow ownership, compact handoffs, and a final report that distinguishes facts, interpretations, comparisons, and forecasts.

Platform notes

• Install this skill under ~/.openclaw/skills/openclaw-deep-research/ for user-wide use or skills/openclaw-deep-research/ inside a workspace when using OpenClaw local skills.
• Keep the actual research directory under the active workspace so report artifacts remain close to the task.
• Assume delegation is optional. Use subagents only if the current OpenClaw setup exposes them and the user explicitly wants parallel work.

Quick start

1. Freeze the topic, audience, report shape, and research cutoff date before collecting sources.
2. Create one canonical project root in the current workspace and treat it as the single source of truth.
3. Persist project memory immediately: status, task board, source ledger, claim ledger, fact-check log, handoffs, and draft report.
4. If the user explicitly wants multi-agent work, split the project into bounded roles with disjoint write scopes. Otherwise use the same workflow locally without delegation.
5. Collect sources first, write claims second, draft only from verified claims, and end with a QA pass that checks dates, comparability, and unsupported conclusions.

Non-negotiables