paper-triage-ranker

name	paper-triage-ranker
description	Rank candidate research papers into must_read, should_read, maybe_read, and skip groups for a paper-to-skill workflow. Use when a coding agent should combine related-paper and innovation candidates, judge relevance, novelty, proof-pattern value, reading cost, access status, and produce a reading_plan.json with user confirmation checkpoints.

Paper Triage Ranker

Purpose

Use this skill to turn candidate paper lists into a reading plan. The output should tell the user what to read first and why, especially for downstream proof-pattern extraction.

Inputs

• candidate_papers.json
• innovation_candidates.json
• Optional research_profile.json
• Optional human_feedback_state.json
• Optional user constraints such as time budget or target research question.

Workflow

1. Load all candidate lists and merge duplicate papers by title, URL, DOI, or arXiv ID.
2. If human_feedback_state.json exists, read it before scoring. Apply focus_updates, negative_preferences, paper_decisions, and next_step_directives.
3. Score each paper for relevance, innovation potential, proof-pattern value, positioning value, reading cost, and access status.
4. Assign each paper to must_read, should_read, maybe_read, or skip.
5. For must_read and should_read, specify what to inspect: abstract, intro, theorem statements, proofs, experiments, related work, or appendices.
6. Mark whether download and Markdown conversion are recommended.
7. Write reading_plan.json using references/schemas.md.
8. Write triage_report.md with a concise reading sequence, applied human feedback, and checkpoint questions.

Required References

Read only as needed:

• references/ranking_rubric.md: scoring and group assignment guidance.
• references/schemas.md: required output fields.

Use the template:

• assets/reading_plan.template.json

Output Files

outputs/<research_project_id>/
├── reading_plan.json
└── triage_report.md

Triage Rules

• Prefer papers that may produce reusable proof-pattern SkillCards.
• Do not let recency dominate if an older paper contains the core theorem or proof method.
• Down-rank papers that are only benchmark or generic background unless positioning requires them.
• Separate reading priority from download status.
• User paper_decisions override default ranking unless doing so would violate access or paywall rules; explain any conflict in triage_report.md.
• Increase the weight of papers matching user focus updates, especially proof-pattern value when the feedback asks for reusable methods.
• Down-rank papers matching negative preferences, such as pure empirical benchmarks, unless needed for positioning.
• Ask for user confirmation before downloading, converting, or extracting from multiple papers.

Completion Check

Before finishing, confirm:

• reading_plan.json is valid JSON.
• Every candidate is assigned to exactly one group.
• must_read and should_read entries include reading focus and extraction recommendation.
• The report includes the user checkpoint for final paper selection.