# Resolve Paper Details for Reliable Reproduction

Repository documentation can omit details that are essential for reproducing deep learning results. This skill resolves narrow gaps from primary paper sources while documenting conflicts.

## Install

```bash
npx skillstore add lllllllama/paper-context-resolver
```

## Metadata

- Status: approved
- Slug: lllllllama-paper-context-resolver
- Skillstore revision: r2
- Version status: missing
- Tree hash: 5c990aae16cbd4d8cd4d70141426edba7f7b099373d0a58f81615aa5da7913fd
- Author: lllllllama
- GitHub username: lllllllama
- License: MIT
- Repository: https://github.com/lllllllama/ai-paper-reproduction-skill/tree/main/skills/paper-context-resolver/
- Ref: 656be3040aef5c047555a908cd5c695d22a4a548
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands
- Quality score: 75
- Public page: https://skillstore.pages.dev/skills/lllllllama-paper-context-resolver
- Manifest: https://skillstore.pages.dev/api/skills/lllllllama-paper-context-resolver/manifest

## Capabilities

- Resolves specific dataset versions and split details from primary paper sources.
- Identifies reproduction-relevant preprocessing and postprocessing requirements.
- Clarifies evaluation protocols that repository documentation leaves ambiguous.
- Maps checkpoints to model variants when source evidence supports the mapping.
- Separates direct source evidence from inference and records README-paper conflicts.

## Use Cases

- Clarify an Evaluation Split: Compare the repository instructions with the linked paper to identify the exact evaluation split and document any disagreement.
- Verify Preprocessing Details: Extract only the preprocessing details needed to complete an experiment when the README omits essential parameters.
- Map Checkpoints to Variants: Use primary source evidence to connect available checkpoints with documented model variants without assuming unsupported mappings.

## Prompt Templates

### Resolve One Missing Detail

```
The README does not specify [missing detail]. Use [paper link] to answer only this reproduction question and distinguish evidence from inference.
```

### Check a Dataset Split

```
Compare the repository evidence with the primary paper for the [dataset] split. State each source's claim and record any conflict.
```

### Verify an Evaluation Protocol

```
Resolve the evaluation protocol for [model and task]. Check metrics, data partition, checkpoint selection, and preprocessing using primary sources only.
```

### Produce an Evidence Resolution

```
Given [repository evidence], [paper sources], and [reproduction question], produce a narrow answer with source claims, conflicts, inferences, and unresolved uncertainty.
```

## Limitations

- Requires a concrete reproduction question and existing repository evidence.
- Does not provide general paper summaries or broad related-work reviews.
- Does not scan repositories, configure environments, or execute commands.
- Cannot resolve details that are absent from the supplied or located primary sources.

## Best Practices

- Start with one concrete reproduction-critical question after reviewing the README and repository files.
- Prefer README-linked papers, official project pages, arXiv, or OpenReview primary records.
- Report source disagreements explicitly and label every inference.

## Anti Patterns

- Do not request a general summary when only one reproduction detail is missing.
- Do not replace README instructions silently with values from a paper.
- Do not present speculative settings as source-supported reproduction requirements.

## Security Audit

- Audited at: 2026-08-08T08:58:52.398\+00:00
- Summary: Both static findings are false positives caused by Markdown backticks around a skill slug and a reference path in SKILL.md. No command execution, prompt injection, or other malicious intent was found.

## Stats

- Views: 98
- Downloads: 31
- Favorites: 0
- Popularity score: 0