deep-research-openclaw-agent_skill

Overview

This skill installs and wires a structured OpenClaw sub-agent called deep-researcher to provide reproducible, artifact-driven deep research. It bundles orchestration, hybrid discovery connectors, an explicit claim ledger, and a lint/validation/finalization pipeline to produce machine-readable final reports. Use it to add a defendable, auditable research workflow to an OpenClaw deployment.

How this skill works

The skill integrates the deep-research-openclaw-agent package into your OpenClaw base by copying the workspace-researcher prompt pack and runtime scripts, registering the deep-researcher agent, and aligning the main-agent handoff contract. During runs the agent follows a plan -> scout -> harvest -> verify -> synthesize pipeline, using hybrid search (web_search, Tavily, web_fetch) when configured, recording sources in a registry, tracking claim coverage, linting reports, and producing a validated M2M JSON finalization with explicit success semantics.

When to use it

• When you need reproducible, auditable deep research runs inside OpenClaw.
• To add hybrid discovery (web + Tavily) to a structured research pipeline.
• When final deliverables must include claim provenance and coverage metrics.
• If you require report linting, validation, and machine-readable finalization.
• For teams that need honest SUCCESS | PARTIAL | FAILURE outcomes with gaps documented.

Best practices

• Ensure OpenClaw 2026.3.x or later is installed and Python is available on the host.
• Copy the workspace-researcher folder into your OpenClaw base or point config at it to avoid path issues.
• Register or update the deep-researcher agent in openclaw.json and align the main handoff contract.
• Provide Tavily API credentials only if you intend to use the Tavily-backed scouting path.
• Run the provided py_compile and init_research_run smoke tests before real tasks to confirm wiring.

Example use cases

• Conducting literature-style investigations with explicit source ledgers and claim verification.
• Running hybrid discovery tasks that combine web_search, Tavily, and direct fetches.
• Producing validated M2M JSON reports for downstream automated tooling.
• Setting up a repeatable research pipeline for auditing, compliance, or archival use.
• Running smoke tests to confirm agent initialization before enabling external research.