---
name: prepare-audit-reproducible-research-package
description: 'Apply a repeatable evidence gate to a research handoff covering provenance, permissions, code, environment, dependencies, seeds, run instructions, expected outputs and reproduction gaps. Use when before handing a computational study to a collaborator, reviewer, repository or archive. When testing whether a replication package contains enough evidence for an independent authorized run. When documenting divergence between expected and observed outputs.'
compatibility: 'Instruction-only Agent Skill. Requires an Agent Skills-compatible client and access to the tools or systems named in the instructions. No executable scripts are bundled.'
metadata:
  publisher: RichlyAI
  product: Amo.ng
  amo_id: AMO-S-000033
  canonical_url: 'https://amo.ng/skills/prepare-audit-reproducible-research-package'
  package_version: 1.0.0
  source_type: prompt
  source_amo_id: AMO-P-000325
  source_content_sha256: 7c62b63e0b7625748ceb7c0c69a8c24483e9ae96639ed044a72df5fc51520b0e
  source_version: 1.2.0
---

# Prepare and Audit a Reproducible Research Package

## When to use this Skill

- Before handing a computational study to a collaborator, reviewer, repository or archive.
- When testing whether a replication package contains enough evidence for an independent authorized run.
- When documenting divergence between expected and observed outputs.

## When not to use this Skill

- To claim successful reproduction without supplied execution records.
- To bypass data licences, participant restrictions, ethics conditions or secure-compute requirements.
- As a substitute for disciplinary methods review, repository acceptance or research-integrity investigation.

## Required inputs

- Research-package manifest and expected output references.
- Data provenance, licences, permissions and access instructions.
- Code, environment specification, dependencies, seeds and run instructions.
- Actual execution logs, output checksums or reviewer observations when a run occurred.
- Known deviations, inaccessible components and decision context.

Use [INPUT-TEMPLATE.md](assets/INPUT-TEMPLATE.md) to organize the supplied information.

## Required tools or access

Published source tool designations: Codex. These are source designations, not portable-client compatibility claims. Use only authorized access described in the inputs and source instructions. Do not claim access or inspection without evidence.

## Procedure

Before starting, read [SOURCE.md](references/SOURCE.md), including the exact Prompt instructions and any ordered Workflow steps, review notes and carry-forward requirements.

Use the source Prompt as the operating audit, not as evidence that execution occurred.

1. Inventory every required package component and bind it to provenance, permission, version and owner evidence.
2. Separate static inspectability from executed reproduction. Record the environment, command, seed, input fingerprint, exit state and output comparison only when corresponding evidence exists.
3. Classify each expected result as reproduced within tolerance, diverged, not tested, inaccessible or not applicable.
4. Trace divergences to the smallest supported hypothesis; preserve alternative explanations and evidence requests.
5. Produce a handoff manifest, reproducibility matrix, deviation register and smallest next action.
6. Assign data-access, licensing, infrastructure, methods and release decisions to their accountable owners.

Never insert credentials or protected data into the Skill input. A package that cannot lawfully be accessed is incomplete, not failed.

Published expected output:

A versioned research-package manifest, execution evidence ledger, expected-versus-observed output matrix, divergence register, access/licensing gap list and bounded reproducibility disposition.

## Review gates

Read and apply the source review instructions in [SOURCE.md](references/SOURCE.md) before acting. Keep proposed, unrun and executed actions distinct; preserve missing evidence and obtain the human approvals required by the source.


## Completion criteria

Complete when the package boundary and versions are fixed; every required artefact has provenance, permission and availability status; every reported run has reproducible evidence; expected outputs have an explicit disposition; divergences and inaccessible elements remain visible; and next actions have accountable owners. Otherwise return an incomplete-package record.

Use the [completion checklist](references/COMPLETION-CHECKLIST.md) and record unresolved items.

## Limitations

- Do not claim an execution, checksum match or reproduced result without a supplied record.
- Do not redistribute licensed, confidential or participant data.
- Do not conceal environment drift or modify the reference package during an audit without a recorded change.
- Repository deposit, ethics, data access and release remain with the responsible owners.

This is an instruction-only package. Packaging does not execute work, grant access or independently verify the package. Existing public verification in amo.json describes only the identified published source resources.

## References

- [Published source instructions](references/SOURCE.md)
- [Completion checklist](references/COMPLETION-CHECKLIST.md)
- [Input template](assets/INPUT-TEMPLATE.md)
- [Rights](RIGHTS.md)
- [Canonical Amo.ng Skill](https://amo.ng/skills/prepare-audit-reproducible-research-package)