The research-harness skill enforces cognitive discipline, structured repository architecture, and strict governance rules for AI-native scientific experimentation and long-running LLM agent collaboration.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install ai-research-harness
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install ai-research-harness using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Research Harness?
The research-harness skill provides a sophisticated cognitive framework designed to guide AI agents through structured, reproducible scientific experimentation. Rather than prescribing specific commands, this Openclaw Skills asset instills a robust mental discipline within AI-native workspaces. It prevents common multi-agent failure points such as premature scaling, unverified overclaiming, silent changes to baseline metrics, or deleting unsuccessful experiment runs.
By establishing systematic constraints, it structures collaborative environments so that research direction remains strictly under human control while execution is fully delegated to agents. Grounded in contemporary multi-layered agent engineering paradigms like ETCLOVG, it introduces essential verification gates that enforce meticulous validation of data pipelines before results are formally interpreted.
Research Harness Use Cases
- Setting up a new AI-native research repository or structured environment for scientific experimentation.
- Designing controlled, multi-group experiments using autonomous LLM agents.
- Running evaluation pipelines requiring rigorous statistics, effect-size calculation, and error analysis.
- Organizing a multi-agent collaborative workspace for long-running, autonomous research tasks.
- Enforcing safe research governance, data provenance tracking, and objective baseline management.
How Research Harness Works
- Scaffold the Environment: The skill initialises a structured three-layer repository consisting of a thinking space for human decisions, an execution layer for agent briefs/logs, and a code workshop for executable artifacts.
- Harden and Contract Validation: Establishes module contracts and local automated validators to enforce an error-free state before formal runs proceed.
- Design Controlled Experiments: Builds attributable experiment groups progressively, ensuring adjacent groups change exactly one variable at a time.
- Execute and Verify: Conducts phased execution loops starting from dry runs and smoke runs up to a full experimental wave.
- Score and Analyze: Implements a dual-track validation system (rule-based and semantic) and computes rigorous statistical parameters like Cohen's d.
- Package and Handoff: Synthesizes results into a claim-safe memo and alignment document to prepare the repository state for subsequent iterations.
Research Harness Setup
To deploy and provision the research environment using this skill, scaffold the directory layout and register core alignment files using the following commands:
# Initialize the three-layer research repository structure
mkdir -p thinking-space execution-layer code-workshop harness scripts
# Generate the essential workspace configuration panels and logs
touch AGENTS.md PLAN.md WORKFLOW.md harness/README.md
# Trigger the initial local repository state validation execution loop
python3 scripts/validate_repo_state.py --mode default
Research Harness Data Schema & Taxonomy
The skill manages research data through a strict structural and metadata schema across the repository layers:
| Location | File / Directory Component | Purpose & Metadata Retention |
|---|---|---|
| Root Workspace | AGENTS.md, PLAN.md |
Workspace configuration mapping, phase execution panel, and workflow records. |
execution-layer/ |
experiments/results/manifest.csv |
Run-level provenance ledger containing run_id, wave, task_id, group, model, and git_commit. |
| Module Level | CONTRACT.md |
Rigid definitions for module purpose, inputs, outputs, invariants, and local validator scripts. |
| Protected Surfaces | baselines/, rubrics/ |
Version-controlled configuration assets containing testing benchmarks, validation rubrics, and raw scoring configurations. |
Research Harness Advanced Features
- Six Cognitive Disciplines: Enforces minimum closed loops, isolated variables, dual-track validation, effect-size prioritisation, pipeline validation, and explicit theoretical grounding.
- Seven Governance Rules: Mandates human-direction ownership, evidence status grading, a strict manifest for failed runs, and protected-surface control.
- ETCLOVG Structural Mapping: Alignable with advanced multi-layer execution, lifecycle, context, and governance taxonomy for enterprise AI agents.
- Methodology Review Gates: Prevents non-trivial or novel design execution without a prior Design Justification Document backed by published scientific precedents.
SKILL.md
Loading
Related Openclaw Skills
Google Maps Extractor (Apify) Skill
A powerful agent skill leveraging Apify's Google Maps Extractor to harvest local business directories, contact information, and detailed place metadata.
Academic Research Skills
An advanced AI-powered academic workflow skill designed to assist researchers in literature synthesis, paper writing, outlines, and systematic reviews.
Academic Research Skills
A comprehensive AI-driven academic workflow skill covering 13 specialized prompts for topic selection, outlining, drafting, reviewing, and answering revisions.
Add Flashcard
A localized spaced repetition flashcard system designed to help users efficiently memorize vocabulary, facts, and key information through active recall.
Search Synthesis Expert
The Search Synthesis Expert skill uses sequential thinking and Playwright browser automation to decompose tasks, execute multi-source searches, and deliver highly credible, synthesized intelligence reports.
Workflow Orchestration Skill
The Workflow Orchestration Skill compiles complex natural language business goals into statically validated Directed Acyclic Graph (DAG) execution plans for downstream engines.
Featured*
