---
name: research-claims
description: Research a topic into a claim ledger with source roles, independence, dates, conflicts, and calibrated confidence. Use when conclusions need current web evidence, including authorized reading from logged-in Chrome pages such as X or Xiaohongshu. Do not use for casual browsing or credential extraction.
license: Apache-2.0
metadata:
  compatibility: Codex, Kimi Code, and Grok Build; supports web search and user-authorized visible Chrome control when available.
  author: Lei Huang
  version: "0.1.0"
  status: experimental
---

# Research Claims

Research is a graph from question to claims to sources, not a pile of links.

## Workflow

1. Define the decision or reader promise, scope, audience, jurisdiction, freshness, and stopping condition.
2. Break the topic into answerable subquestions and candidate claims.
3. Search source-first. Prefer official documents, standards, papers, code, filings, and first-party announcements for exact facts.
4. Use named experts for their experience or positions and reputable secondary sources for synthesis.
5. Treat viral or anonymous content as discovery leads until verified. Use social posts as primary evidence only for what the author said or when discourse is the subject.
6. Read the source, not only the search snippet. Record publication date, access date, and material limitations.
7. Trace dependence: several articles repeating one original report are one source chain.
8. Seek disconfirming evidence and record conflicts beside the claim.
9. Write the synthesis from the ledger. Attach citations directly to supported claims.
10. Audit every material claim for direct support and calibrated wording.

## Existing authenticated Chrome

Use the user’s existing Chrome state only when the user authorizes browser-based research or the task explicitly requires a logged-in source.

Allowed:

- navigate visible pages, search, click, scroll, and read rendered content;
- use an already logged-in X, Xiaohongshu, or other site session;
- preserve page URL, author, publication date, access timestamp, and visible context;
- ask the user to take control for login, CAPTCHA, consent, or access decisions.

Prohibited:

- inspect, export, copy, or reveal cookies, passwords, browser profiles, local storage, session tokens, or authentication headers;
- bypass CAPTCHA, paywalls, consent, robots policy, rate limits, or access controls;
- run hidden bulk scraping or continue after the site blocks access;
- publish, like, follow, message, or modify account state unless separately and explicitly authorized.

## Claim ledger

Write `.workshop/research/<slug>/claims.csv` or an equivalent table with:

```text
claim_id,claim,source_author,url,published_at,accessed_at,source_role,independent,support,confidence,evidence_note,conflict_or_limit
```

Allowed support values: `direct`, `partial`, `conflicting`, `contextual`. Confidence is `high`, `medium`, or `low`.

## Output

Write a brief containing executive answer, claim-by-claim synthesis, counterevidence, practical implications, open questions, method, and source list. The evidence packet reports inaccessible sources, removed or downgraded claims, authenticated-browser interventions, and freshness limits.