snapshot-writer_skill

Overview

This skill produces a one-page, bullets-first literature snapshot (output/SNAPSHOT.md) from a small core set of papers and a concise outline. It prioritizes high-signal claims, concrete reading pointers, and contrastive evaluation anchors rather than long narrative surveys. The output is audit-friendly: pointers must map to papers/core_set.csv and avoid invented citations.

How this skill works

It reads outline/outline.yml to find the topic boundary and write-worthy bullets, then ingests papers/core_set.csv to build a pointer palette of canonical anchors, strong baselines, and benchmarks. The skill writes a compact SNAPSHOT.md with a fixed structure: title+scope, evidence policy, taxonomy, key themes (claim -> why it matters -> pointers), reading path, and open problems. It enforces guardrails: no invented papers, minimal prose, and pointers in the stable P#### format.

When to use it

• You need a readable 24–48h research snapshot, not a full evidence-first survey.
• You want a bullet-first, actionable reading path with concrete paper pointers.
• You have a small curated core set (papers/core_set.csv) and an outline/outline.yml.
• You need to hand off a high-signal summary for rapid decisions or literature triage.
• You must avoid long-form narrative and prefer auditability of citations.

Best practices

• Keep bullets content-bearing: claim -> why it matters -> pointer(s).
• Select a mix of canonical anchors, recent strong baselines, and benchmark papers.
• Make at least two cross-paper contrasts to provide evaluative anchors.
• If evidence is abstract-only, state that once in the evidence policy bullet.
• Limit prose to short paragraphs (<= 3 lines) and prefer bullets for all claims.

Example use cases

• Produce a one-page snapshot to brief collaborators before a kickoff meeting.
• Create a reading path for interns to prioritize the 10–15 must-read papers.
• Deliver a compact literature overview for grant pitches or project proposals.
• Quickly compare two competing approaches with contrastive bullets and pointers.
• Generate a benchmark/coverage bullet that maps datasets and protocols to papers.