agentsclimarketplace

Research harness release

Skill maxwell2732/claudecode-research-harness-workflow/skills/research-harness-release

RES: Build the replication/evidence package after a review APPROVE verdict. Assembles scripts, logs, cleaned-data docs, tables, figures, and reproducibility report. Requires human confirmation gate. Trigger: release research, build replication package, package evidence, finalize study. Do NOT load for: planning, execution, review, setup, audit, cleaning.From its SKILL.md

Install
npx -y skills add maxwell2732/claudecode-research-harness-workflow --skill research-harness-release

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

9.3 KB, as published. Nobody here has run it

Research Harness Release

Assemble the complete replication/evidence package after a successful review. This skill requires an APPROVE verdict from /research-harness-review before it will package any output.

A single human confirmation gate precedes all packaging operations.

This skill runs after /research-harness-review (verdict: APPROVE).

Quick Reference

InputAction
/research-harness-releaseFull replication package assembly
/research-harness-release --dry-runList what would be included; do not copy or create the package
/research-harness-release --no-dataPackage without processed data (for confidential data projects)

Pre-flight Checks

  1. Read reports/review_report.md. If it does not exist, stop.
  2. Check the verdict in reports/review_report.md. If it is not APPROVE, stop and tell the user: the review must pass before release can proceed.
  3. Read study_spec.md and analysis_plan.md. If either does not exist, stop.
  4. Confirm that all tasks marked cc:done in analysis_plan.md have a corresponding evidence row (script + log + output).

If the review verdict is REQUEST_CHANGES or BLOCK: do not proceed. Tell the user to resolve review findings first.

Procedure

Step 1 — Inventory all artifacts

Collect paths for every artifact that will go into the replication package:

Source-of-truth files:

  • study_spec.md
  • analysis_plan.md

Reports:

  • reports/data_audit_report.md
  • reports/data_cleaning_plan.md (if exists)
  • reports/data_cleaning_report.md
  • reports/merge_report.md (if merges occurred)
  • reports/review_report.md

Scripts: all scripts referenced in analysis_plan.md Evidence Ledger

Logs: all logs referenced in analysis_plan.md Evidence Ledger

Outputs: all tables and figures referenced in analysis_plan.md Evidence Ledger

Cleaned data: data/processed/ files referenced in the cleaning report (skip if --no-data)

Templates used: list only (do not copy templates themselves)

For each artifact, record whether the file exists on disk. Flag missing files as gaps.

Step 2 — Human confirmation gate

Print the full artifact inventory:

Research Harness Release — Confirmation Required

Review verdict: APPROVE
Review report: reports/review_report.md

Artifacts to include:

  Source-of-truth:
    [x] study_spec.md
    [x] analysis_plan.md

  Reports:
    [x] reports/data_audit_report.md
    [x] reports/data_cleaning_report.md
    [x] reports/merge_report.md
    [x] reports/review_report.md

  Scripts (N files):
    [x] scripts/clean.R
    [x] analysis/main/scripts/main_regression.R
    ...

  Logs (N files):
    [x] logs/clean_YYYYMMDD.log
    [x] analysis/main/logs/main_regression.log
    ...

  Outputs (N files):
    [x] output/tables/table1.csv
    [x] output/tables/table2.csv
    [x] output/figures/figure1.pdf
    ...

  Cleaned data:
    [x] data/processed/analysis_ready.csv    [included / excluded with --no-data]

Gaps (files listed in analysis_plan.md but not found on disk):
  [list or "none"]

Data access note:
  [If --no-data: "Processed data excluded. Add access instructions to reproducibility_report.md."]
  [If data included: "Processed data included. Confirm you have permission to share this data."]

Proceed with release? (yes / no)

Wait for user confirmation. Do not proceed without it.

If the user says no: stop. Print: "Release cancelled. No files were packaged."

Step 3 — Write reproducibility_report.md

Copy templates/reproducibility_report.md to reports/reproducibility_report.md.

Fill in all sections:

§1 Software environment: read from script headers (language and package versions mentioned in comments or sessionInfo() / pip freeze output in logs if present; otherwise write unknown)

§2 Scripts: list all scripts from the Evidence Ledger in execution order. Record file name and last-modified timestamp (SHA if git is available).

§3 Data files: list raw data files (with access instructions) and processed data files (with paths). If --no-data, note that processed data is not included and explain how to obtain raw data.

§4 Logs: list all logs from the Evidence Ledger with date run and exit code (read from log header).

§5 Output files: list all tables and figures from the Evidence Ledger.

§6 Verified claims: for each key numerical claim verified during review (from reports/review_report.md §3), list the claim, the reported value, and the log file where it appears.

§7 Unverified items: list any items from reports/review_report.md that were flagged as unverified or advisory, with risk level.

§8 How to reproduce: write step-by-step instructions. Steps must be specific (e.g., "Run Rscript scripts/clean.R from the project root") not generic.

§9 Remaining limitations: copy from reports/review_report.md minor findings and from study_spec.md §7 open questions.

§10 Checklist: fill in each item as checked or explain why it is not applicable.

Step 4 — Create the release folder

Create release/ at the project root (or the user-specified path).

Copy all artifacts from Step 1 into release/ preserving the folder structure:

release/
├── study_spec.md
├── analysis_plan.md
├── reports/
│   ├── data_audit_report.md
│   ├── data_cleaning_report.md
│   ├── merge_report.md          (if applicable)
│   └── review_report.md
├── scripts/
│   └── clean.R
├── analysis/
│   └── <stage>/scripts/<task>.R
├── logs/
│   └── clean_YYYYMMDD.log
├── analysis/
│   └── <stage>/logs/<task>.log
├── output/
│   ├── tables/
│   └── figures/
├── data/processed/              (omit if --no-data)
└── reports/reproducibility_report.md

Do not include raw data under data/raw/ unless the user explicitly instructs it (and it is not confidential).

Step 5 — Final verification

After copying:

  • Confirm every file in the inventory exists in release/
  • Re-read the reproducibility report checklist — confirm all items are checked
  • Count: total files in package, total verified claims, total unverified items

Step 6 — Print release summary

Research Harness Release — Complete

Package location: release/
Total files: N
Verified claims: N
Unverified items: N (see reports/reproducibility_report.md §7)

⚠ Before sharing this package:
  - Confirm you have permission to share any data files included
  - Confirm data/raw/ contents are not in release/ (unless intended)
  - Test reproduction instructions on a clean environment if possible

Reproducibility report: release/reports/reproducibility_report.md

Forbidden Actions

  • Do not proceed if the review verdict is not APPROVE
  • Do not proceed without human confirmation (Step 2)
  • Do not include files from data/raw/ unless the user explicitly approves
  • Do not claim full reproducibility if data access restrictions prevent it — document them in §3 and §8
  • Do not omit known limitations from the reproducibility report
  • Do not fabricate software version numbers or timestamps not present in logs

Evidence Requirements

  • reports/review_report.md with verdict APPROVE
  • reports/reproducibility_report.md with all sections filled in
  • release/ folder exists and contains all inventoried artifacts
  • Release summary printed

Completion Criteria

  • Review verdict confirmed as APPROVE
  • Human confirmation received
  • Artifact inventory complete with no unexpected gaps
  • reports/reproducibility_report.md written with all sections
  • release/ folder created and populated
  • Reproducibility report checklist fully checked
  • Release summary printed with file count and verified claims

Post-Release

The release package is complete. Remaining researcher responsibilities (outside harness scope):

  • Submit the package to a data repository (e.g., Harvard Dataverse, ICPSR, OSF)
  • Add a README to the repository explaining how to obtain restricted data (if applicable)
  • Archive the release/ folder alongside the paper submission

Keep looking

Skills are one crate of 614,496. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.