# Paperpilot

> Resolver-first skill for academic paper reading and analysis. Use when the user request to read/analyze a paper with a certain resource (uploaded files or online URLs/DOI/arXiv links), classify survey vs non-survey, and execute the correct reading workflow. Default to first-principles detailed reading for non-survey papers and survey-reading workflow for survey papers. Use evaluate-rubric as the standard only when the user explicitly asks to evaluate/critique the paper.

- Skill: `eyixinwang/paperpilot` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add eyixinwang/paperpilot`
- Raw SKILL.md: https://api.skillmd.com/api/skills/eyixinwang/paperpilot/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: eyixinwang (https://skillmd.com/u/eyixinwang)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/eyixinwang/paperpilot

---


# PaperPilot

## Overview

Use this skill to analyze papers from recognized content only. Follow a minimal 3-step pipeline: resolve resources, classify paper type, then read/analyze.

## Mandatory Rule

- Do not synthesize from memory or unresolved sources.
- Start analysis only after at least one paper resource is concretely resolved and recognized.

## Workflow

### Step 1: Confirm concrete resources and resolve with built-in tools

Resolve user-provided resources with built-in tools:

- uploaded artifacts: built-in file/PDF readers
- online artifacts: built-in URL resolvers for URL/DOI/arXiv/publisher links

Follow [resource-resolution-protocol.md](references/resource-resolution-protocol.md). For each source, record:

- source id and input type
- retrieval status (`resolved`, `partial`, `failed`)
- resolved artifact type (`pdf`, `html`, `txt`)
- recognized content volume (rough character count)
- blocker notes

If all sources are unresolved, stop and request corrected or additional resources.

### Step 2: Classify paper type and route workflow

Use [paper-type-routing.md](references/paper-type-routing.md) to classify:

- `survey`
- `non-survey`
- `hybrid/uncertain`

Record:

- routed type
- confidence (`high`, `medium`, `low`)
- evidence anchors used for routing

Route:

- `survey` -> survey synthesis workflow
- `non-survey` -> first-principles detailed reading workflow (default)
- `hybrid/uncertain` -> run both with explicit uncertainty note

### Step 3: Read and analyze

For `non-survey` papers, default to first-principles detailed reading using [first-principles-reading-workflow.md](references/first-principles-reading-workflow.md), including the strict 8-point.

For `survey` papers, default to [survey-reading-workflow.md](references/survey-reading-workflow.md), including the strict 7-part sequence.

If the user requests structured presentation, include method taxonomy using a table or hierarchical structure.

If the user explicitly asks to evaluate, critique, assess quality, or review the paper, apply [evaluate-rubric.md](references/evaluate-rubric.md) as the scoring standard.

Output style requirements:

- direct technical writing, no pleasantries
- explicit evidence anchors for major claims
- clear separation of facts vs inferences
- explicit uncertainty marking when evidence is partial

## Resources

### `references/resource-resolution-protocol.md`

Resolver-first constraints and source-resolution requirements.

### `references/paper-type-routing.md`

Survey vs non-survey classification signals and routing requirements.

### `references/first-principles-reading-workflow.md`

Default deep-reading method for non-survey papers.

### `references/evaluate-rubric.md`

Evaluation standard to use only when the user requests paper evaluation/critique.

### `references/survey-reading-workflow.md`

Default deep-reading method for survey/review papers.