---
name: research-with-fullsy
description: Research a web question or competitor set with verifiable Fullsy receipts. Use when the user wants sourced web research, competitor research, fact-finding, or a memo where each materially relied-on source should have a Fullsy Capture Record. Do not use when Codex lacks JavaScript access to the user's Chrome profile with Fullsy Agent Access enabled.
---

# Research with Fullsy

Research normally, but leave a Fullsy receipt for every material web source used in the answer.

## Preconditions

- Fullsy must be installed in the Chrome profile being automated.
- Fullsy Agent Access must be enabled and the control-page origin approved.
- You must have a browser tool that can evaluate JavaScript in that same Chrome profile.
- If any precondition is missing, say exactly what is missing and stop the Fullsy workflow. Do not pretend a normal HTTP fetch created a record.

## Fullsy bridge

Run calls from one trusted, already-approved control tab. The API is:

```js
await window.fullsy.captureUrl(url, { caseName })
await window.fullsy.capture({ caseName })
await window.fullsy.getRecord(capId)
await window.fullsy.list(limit)
```

`captureUrl` and `capture` return quickly with `{ ok, capId, status: 'started' }`. Poll `getRecord(capId)` until `ok:true` before treating the source as captured.

Fullsy permits at most one agent capture every 15 seconds. Serialize captures. If rate-limited, wait and retry; do not create parallel capture storms.

## Workflow

1. Restate the research question internally and define what evidence would answer it.
2. Create a concise case name that describes the errand, such as `CRM pricing comparison 2026-08-17`.
3. Find the strongest sources for the question. Prefer primary sources for factual claims when available.
4. Read a source before deciding to rely on it.
5. For every source materially used in the final answer, capture its canonical HTTP(S) URL with `window.fullsy.captureUrl(url, { caseName })`.
6. Poll `getRecord(capId)` until the record exists. Save its record id, captured URL, title, capture time, case name, and whether an integrity manifest is present.
7. If capture fails, either retry safely or exclude the source from claims that require a Fullsy receipt. Never invent a record id.
8. Synthesize the answer from the evidence. Separate observed facts from inference.
9. If sources conflict, show the conflict instead of forcing false consensus.

## Citation discipline

For each material factual claim, retain both:

- the ordinary source URL/citation used by the research environment; and
- the corresponding Fullsy record id.

A Fullsy fingerprint proves that the local captured artifact has not changed since capture. It does not prove that the publisher's statement was truthful.

## Final output

Use this structure unless the user requests another format:

### Answer
Concise synthesis of the research.

### Evidence ledger
| Claim / source | Source | Fullsy record | Captured | Status |
|---|---|---:|---|---|

Use `fingerprinted` when an integrity manifest exists. Do not call it `verified` unless the workflow actually re-verified the stored record.

### Uncertainty
List unresolved conflicts, assumptions, stale data risks, or sources that could not be captured.