# Future-House--edison-client--owl-precedent-search

> Owl — Scientific Precedent Search (FutureHouse Platform)

- Skill: `qhjqhj00/future-house-edison-client-owl-precedent-search-2` (Agent Skill)
- Install (CLI): `npx skillmds@latest add qhjqhj00/future-house-edison-client-owl-precedent-search-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/qhjqhj00/future-house-edison-client-owl-precedent-search-2/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: qhjqhj00 (https://skillmd.com/u/qhjqhj00)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/qhjqhj00/future-house-edison-client-owl-precedent-search-2

---


# Owl — Scientific Precedent Search (FutureHouse Platform)

Owl is FutureHouse's "has-anyone-ever-done-X" agent (job name `PRECEDENT`, formerly known as HasAnyone). Where Crow answers *what is known* and Falcon synthesizes *what the field thinks*, Owl specifically asks **whether a specific thing has ever been reported**.

Use this skill when a researcher is about to start an experiment / build a tool / pursue a hypothesis and wants to confirm it's actually novel — and when not, to find the closest prior work.

## Prerequisites

- `pip install edison-client`
- `EDISON_API_KEY` from <https://platform.edisonscientific.com/profile>

## Minimal usage

```python
import os
from edison_client import EdisonClient, JobNames

client = EdisonClient(api_key=os.environ["EDISON_API_KEY"])
resp = client.run_tasks_until_done({
    "name": JobNames.PRECEDENT,
    "query": "Has anyone tested therapeutic exerkines in humans or non-human primates?",
})
print(resp.formatted_answer)
```

The answer is phrased as a precedent verdict — "Yes, see X et al. 2023 …" or "No published precedent found, but the closest work is Y et al. 2020 which …" — with citations.

## How to phrase queries

Owl works best with **specific** "has anyone" framings. Compare:

| Bad (too broad — use Falcon) | Good (precedent question) |
|---|---|
| "Tell me about exerkines" | "Has anyone delivered exerkines therapeutically in NHPs?" |
| "What's known about CAR-T?" | "Has anyone published a CAR-T construct targeting GPC3 in pediatric solid tumors?" |
| "Review optical clocks" | "Has anyone built a portable strontium optical clock under 50 kg?" |

The agent expects the question to *imply a precedent test*. If the user's phrasing doesn't, rewrite it before sending.

## Recipes

### Pre-experiment novelty check
```python
hypothesis = "Knock-in of a humanized FOXP3 allele into NSG mice rescues Treg development."
resp = client.run_tasks_until_done({
    "name": JobNames.PRECEDENT,
    "query": f"Has anyone tested whether {hypothesis}",
})
if resp.has_successful_answer:
    print(resp.formatted_answer)
else:
    print("Owl couldn't determine — try Falcon for a broader synthesis.")
```

### Prior-art check before patent / paper draft
Run Owl on the central claim in a methods section to confirm it's actually new.

### Finding the closest precedent (when answer is "yes")
Owl always returns the closest matching paper(s). Use those as a starting point for your related-work review. To go deeper, follow up with Crow on the specific paper:

```python
prev_id = client.create_task({
    "name": JobNames.PRECEDENT,
    "query": "Has anyone used base editing to introduce SCD1 mutations in human iPSCs?",
})
follow = client.run_tasks_until_done({
    "name": JobNames.LITERATURE,
    "query": "What were the editing efficiencies and off-target rates reported in the prior work you found?",
    "runtime_config": {"continued_job_id": prev_id},
})
```

## Cost / latency

- ~30 s – 2 min per call
- Credits: similar to Crow (cheaper than Falcon)

## Failure modes

- **`has_successful_answer == False`** with answer "I couldn't find a clear precedent": treat as soft evidence of novelty, but it could also mean the question wasn't specific enough. Consider asking the user to refine.
- Vague "tell me about X" queries waste credits — rewrite into a "has anyone …" form before sending.