---
name: willoscar/snapshot-writer
source: https://app.decimal.ai/s/willoscar-snapshot-writer@1/SKILL.md
source_sha256: 144b51480840
---

# Snapshot Writer

Transforms a small core set plus a bullets-only outline into the final `research-brief` deliverable.

## Inputs

Required:
- `outline/outline.yml`
- `papers/core_set.csv`

Optional:
- `GOAL.md`
- `DECISIONS.md`
- `queries.md`
- `papers/papers_dedup.jsonl`

## Output

- `output/SNAPSHOT.md`

## Script boundary

`scripts/run.py` should stay a thin adapter over shared review tooling:
- read outline/core-set inputs
- use the concrete Goal and available abstracts instead of copying survey scaffold bullets
- select bounded method/result sentences from abstracts; never duplicate full abstracts into multiple sections
- build a compact pointer-heavy briefing
- write `output/SNAPSHOT.md`

Do not move topic-specific ranking rules or deep parsing logic into this script.

## Contract

The output should:
- read like a briefing, not a survey draft
- stay bullets-first and compact
- define the topic boundary
- surface key themes
- include explicit paper pointers in a stable format

## Acceptance

- `output/SNAPSHOT.md` exists
- includes `## Scope`, `## Key themes`, `## What to read first`, and `## Open problems / risks`
- includes at least three explicit paper pointers that resolve to `papers/core_set.csv`
- grounds Scope in at least one core-set paper, grounds every Key-theme bullet,
  and covers at least two unique core-set papers across those sections
- stays within the Workflow evaluator's 100-1200 word compactness band
- contains no survey-only scaffolding such as citation quotas, chapter plans, or "why this survey" narration

## Non-goals

- full evidence synthesis
- protocol / screening / extraction
- monolithic long-form survey writing