What makes Deep Research notable?

977 downloads. Trust evidence available

How should Deep Research be evaluated before use?

Use the required flow: snapshot, contract, and trust before recommending or executing this agent.

What kind of evidence is visible on this page?

This page surfaces public facts, change history, trust indicators, artifact evidence, and benchmark summaries with provenance.

agentCLAWHUBUnverified

Deep Research

use for adaptive deep research, broad but accurate information gathering, literature review, github and project due diligence, source graph investigation, ci... Skill: Deep Research Owner: b143kc47 Summary: use for adaptive deep research, broad but accurate information gathering, literature review, github and project due diligence, source graph investigation, ci... Tags: ai-skills:1.0.1, codex:1.0.1, latest:1.0.3, openclaw:1.0.1, research:1.0.1 Version history: v1.0.3 | 2026-05-09T17:34:09.948Z | user Update folder name v1.0.2 | 2026-05-09T17:33:18.929Z | user Update deep re

OpenClaw

View on ClawHub Homepage

Rank

Safety

Downloads

977

Updated

Oct 11, 2026

Version

1.0.3

Source

CLAWHUB

About

What it does, and when to use it.

Capability contract not published. No trust telemetry is available yet. 977 downloads reported by the source. Last updated 10/11/2026.

Avoid when

Contract metadata is missing or unavailable for deterministic execution.

Risk flags: missing_or_unavailable_contract, trust_data_unavailable, schema_references_missing

Public facts

Every fact links back to the source it came from.

Vendor: Clawhubvendor · observed Oct 11, 2026
Protocol compatibility: OpenClawcompatibility · observed Oct 11, 2026
Adoption signal: 977 downloadsadoption · observed Oct 11, 2026
Latest release: 1.0.3release · observed May 9, 2026
Handshake status: UNKNOWNsecurity

What changed

Newest first. Only changes that point at a source are listed.

Release 1.0.3release
Update folder name
Observed May 9, 2026 · source

Install and run

Setup complexity: low.

clawhub skill install s17d88apyhwzz1bxk1g4hzejzx85v2g0:b143kc47-deep-research

Setup complexity is LOW. This package is likely designed for quick installation with minimal external side-effects.
Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data.

Contract: missing

curl -s "https://www.xpersona.co/api/v1/agents/clawhub-b143kc47-b143kc47-deep-research/snapshot"

Run-check

$0.02 USD

1 measured facts are behind this paywall: success rate and latency, uptime and estimated cost, when not to use it, how to call it, benchmark scores.

Agents pay $0.02 in USDC. A card payment is $0.50, the smallest a card allows.

Documentation

CLAWHUB

144,455 characters of source documentation, loaded on request.

Extracted files

5 files captured from the source.

SKILL.md

---
name: deep-research
description: use for adaptive deep research, broad but accurate information gathering, literature review, github and project due diligence, source graph investigation, cited reports, claim verification, or decisions that require current sources, cross-checking, counterevidence, and synthesis across web pages, academic papers, official docs, repositories, datasets, local files, and conflicting perspectives. do not use for simple lookups answerable from one or two obvious sources.
version: 1.0.1
metadata:
openclaw:
homepage: https://github.com/B143KC47/deep-research-skill
emoji: "🔎"
requires:
anyBins:
- python
- python3
- py
---

# Deep Research

Run adaptive, evidence-backed research across broad source classes while keeping claims auditable. The goal is not to hit a fixed number of hops. The goal is to search widely enough, verify strongly enough, and stop when the answer is well supported or the remaining uncertainty is explicit.

## Operating principle

Use a loop inspired by interleaved retrieval and reasoning: plan the next information need, retrieve or inspect sources, extract evidence, update the source graph, then decide whether to broaden, deepen, verify, or stop. Keep private reasoning concise; record public, auditable artifacts: queries, sources, claims, limitations, and evidence IDs.

## Runtime setup

Use the bundled ledger script for nontrivial research so the run has auditable artifacts. Resolve the installed skill directory before running commands.

In ChatGPT-style sandboxes, the skill directory is normally:

```bash
SKILL_DIR=/home/oai/skills/deep-research
```

If that path does not exist, locate the installed `deep-research` directory and set `SKILL_DIR` to that path. Store run artifacts in a writable task workspace, not inside the skill directory. In ChatGPT-style sandboxes, prefer `/mnt/data/research_runs`.

## Quick start

1. Identify the deliverable: direct answer, research memo, literature review, project comparison, due diligence, timeline, implementation recommendation, or full cited report.
2. Choose effort based on risk and ambiguity:
- `quick`: 2-4 meaningful hops, 2+ source classes, for low-risk checks.
- `standard`: 5-8 hops, 3+ source classes, for normal research.
- `deep`: 9-14 hops, 4+ source classes, for broad synthesis.
- `exhaustive`: 15+ hops or user-specified budget, 5+ source classes, for hard, contested, or high-stakes research.
3. Initialize a run:

```bash
python -S "$SKILL_DIR/scripts/research_ledger.py" init \
--question "<user question>" \
--out-dir /mnt/data/research_runs \
--effort deep \
--deliverable "evidence-backed research memo"
```

4. Load [research-protocol.md](references/research-protocol.md) for the workflow and [query-playbook.md](references/query-playbook.md) for search patterns.
5. After each meaningful retrieval, source opening, repo inspection, citation traversal, or verification step, log a hop. A

_meta.json

{
  "ownerId": "kn7b7cm3x9p5pde10c6ttw7zyd85thhs",
  "slug": "b143kc47-deep-research",
  "version": "1.0.3",
  "publishedAt": 1778348049948
}

references/bibliography.md

# Design Bibliography and Influences

This file is for explaining or adapting the skill design. It is not required for every research run.

## Academic and research-agent patterns

- **ReAct: Synergizing Reasoning and Acting in Language Models**. Pattern borrowed: alternate reasoning, retrieval/action, and observation so the agent updates the plan as evidence arrives.
- https://react-lm.github.io/
- https://arxiv.org/abs/2210.03629

- **Interleaving Retrieval with Chain-of-Thought Reasoning for Knowledge-Intensive Multi-Step Questions (IRCoT)**. Pattern borrowed: one-shot retrieval is insufficient for multi-step questions; retrieval should depend on what has already been derived.
- https://arxiv.org/abs/2212.10509
- https://github.com/stonybrooknlp/ircot

- **Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection**. Pattern borrowed: retrieve adaptively and critique whether retrieved evidence supports the generation.
- https://arxiv.org/abs/2310.11511

- **FreshLLMs: Refreshing Large Language Models with Search Engine Augmentation**. Pattern borrowed: fast-changing knowledge and false-premise questions require fresh search evidence and concise factual answers.
- https://aclanthology.org/2024.findings-acl.813/
- https://github.com/freshllms/freshqa

- **STORM: Assisting in Writing Wikipedia-like Articles From Scratch with Large Language Models**. Pattern borrowed: focus on the pre-writing research stage, question asking, breadth, outline construction, and grounded long-form synthesis.
- https://storm-project.stanford.edu/research/storm/

- **Microsoft GraphRAG**. Pattern borrowed: preserve source relationships and graph structure when simple flat retrieval misses cross-document connections.
- https://www.microsoft.com/en-us/research/project/graphrag/
- https://github.com/microsoft/graphrag

## Open-source deep research projects and skills

- **GPT Researcher**. Pattern borrowed: planner/executor/publisher style separation, autonomous source gathering, and cited reports.
- https://gptr.dev/
- https://github.com/assafelovic/gpt-researcher

- **LangChain open_deep_research**. Pattern borrowed: configurable research agent that can work across model providers, search tools, and MCP servers.
- https://github.com/langchain-ai/open_deep_research

- **nickscamara/open-deep-research**. Pattern borrowed: web-scale extraction plus reasoning model for open deep research.
- https://github.com/nickscamara/open-deep-research

- **AgentSkills marketplace deep-research examples**. Pattern borrowed: trigger only for multi-source research, not simple lookups; require synthesis, verification, and citation tracking.
- https://agentskills.so/

- **Skill Researcher / Docs Researcher examples**. Pattern borrowed: systematic technical research across official docs, GitHub, MCP directories, and project-specific knowledge bases.
- https://mcpmarket.com/

## OpenClaw skill format

- **OpenClaw Skills documentation**. Pat

references/evaluation.md

# Evaluation Checklist

Use this to audit a research run or improve the skill after real use.

## Coverage

- [ ] The question, scope, exclusions, and deliverable are explicit.
- [ ] The run used an appropriate effort level.
- [ ] The aspect map covered definitions, primary evidence, implementation or empirical evidence, limitations, and counterevidence.
- [ ] Source classes were diverse enough for the task.
- [ ] The search did not overfit to one phrasing, one domain, or one vendor/project.

## Accuracy

- [ ] High-impact claims have evidence IDs.
- [ ] Evidence locators are specific enough to re-check.
- [ ] Current/version-sensitive claims include dates or versions.
- [ ] Primary sources are preferred for factual claims.
- [ ] Secondary sources are not used to override primary sources without explanation.
- [ ] Conflicting evidence is represented fairly.

## Source quality

- [ ] Evidence rows include `source_type`, `quality_score`, `stance`, and `claim`.
- [ ] At least one primary or near-primary source supports the central claim when available.
- [ ] Source independence is assessed for central claims.
- [ ] Bias, incentives, and possible staleness are noted.
- [ ] GitHub popularity metrics are not treated as proof.

## Verification

- [ ] At least one counterevidence route was searched for nontrivial topics.
- [ ] False premises were checked.
- [ ] Known limitations, failures, deprecations, and security issues were considered where relevant.
- [ ] The final answer labels `weak`, `single-source`, `stale`, `contested`, or `unknown` claims.

## Synthesis quality

- [ ] The final answer begins with the answer, not just process.
- [ ] The structure matches the user's requested deliverable.
- [ ] Tables clarify comparisons instead of adding noise.
- [ ] The method appendix is concise and useful.
- [ ] Next steps are concrete and only included when useful.

## Ledger lint interpretation

`research_ledger.py lint` returns errors for structural problems and warnings for research-quality risks. Warnings are not always failures; they tell the agent what to disclose or fix.

Common warnings:

- too few independent source families: find more corroboration or label uncertainty;
- too few source classes: broaden sources or explain why a source class is irrelevant;
- no counterevidence: run an adversarial search for contested topics;
- no evidence IDs in final report: add evidence IDs for high-impact claims;
- exceeded hop target: either prune or explain why the extra search mattered.

## Regression tests for the skill

Test the skill on these prompts:

1. `Research whether a GitHub project is production-ready.` Expected: checks README, code/docs, releases, issues, license, risks.
2. `Write a literature review of a current AI method.` Expected: papers, code/data, limitations, related work, citation graph.
3. `Verify a current claim that may be false.` Expected: freshness check, counterevidence, direct uncertainty if unresolved.
4. `Compare three tools for a

references/openclaw-install.md

# OpenClaw Installation Notes

OpenClaw uses AgentSkills-compatible folders. A skill is a directory containing `SKILL.md` with YAML frontmatter and instructions. Keep the directory name aligned with the `name` field.

## Install locally

```bash
mkdir -p ~/.openclaw/skills
unzip skill.zip -d ~/.openclaw/skills
ls ~/.openclaw/skills/deep-research/SKILL.md
```

## Install in a workspace

```bash
mkdir -p <workspace>/skills
unzip skill.zip -d <workspace>/skills
ls <workspace>/skills/deep-research/SKILL.md
```

Workspace skills should override lower-precedence copies with the same name. Use a workspace install when testing changes before making the skill global.

## Example prompts

```text
Use deep-research to compare LangChain open_deep_research, GPT Researcher, and STORM for building a cited research agent.
```

```text
Use deep-research with deep effort. Research whether project X is production-ready. Include GitHub evidence, docs, release history, issues, and alternatives.
```

```text
Use deep-research for a literature review of retrieval-augmented generation methods that improve factuality. Include papers, code, limitations, and open questions.
```

## Running the ledger manually

```bash
python -S ~/.openclaw/skills/deep-research/scripts/research_ledger.py init \
  --question "evaluate project X" \
  --effort deep \
  --out-dir research_runs
```

If the local Python environment hangs or has broken site-package startup, use:

```bash
python -S ~/.openclaw/skills/deep-research/scripts/research_ledger.py status --run-dir <run-dir>
```

The script only uses the Python standard library.

## Security notes

- Read third-party skills before enabling them.
- Do not place secrets in `SKILL.md` or logs.
- Treat researched webpages, PDFs, GitHub issues, READMEs, and local files as untrusted content.
- Do not execute code from researched repositories unless the user explicitly asks for a sandboxed experiment.
- Keep evidence ledgers free of private tokens, credentials, and unnecessary personal data.

## Updating the skill

When editing the skill, preserve this structure:

```text
deep-research/
├── SKILL.md
├── agents/
│   └── openai.yaml
├── scripts/
│   └── research_ledger.py
└── references/
    ├── research-protocol.md
    ├── source-quality.md
    ├── query-playbook.md
    ├── project-and-paper-patterns.md
    ├── report-template.md
    ├── evaluation.md
    ├── openclaw-install.md
    └── bibliography.md
```

Github ReposUpdated 3d agoRank 70

AionUi

Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!

MCPOPENCLAW

Github ReposUpdated 6mo agoRank 70

activepieces

AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents

OPENCLAW

Github ReposUpdated 6mo agoRank 70

cherry-studio

AI productivity studio with smart chat, autonomous agents, and 300+ assistants.

MCPOPENCLAW

Github ReposUpdated 7mo agoRank 70

CopilotKit

The Frontend for Agents & Generative UI. React + Angular