mcpField
D
develop-contribution

develop-contribution

github.com/vincenzoimpProductivity466 installsUpdated 14w ago
Repository

About develop-contribution

Develop Contribution is a skill for creating or regularizing research contribution folders in a project. It guides evidence planning, reproducibility, ethical data handling, and report generation so each contribution is self-contained and documented.

Key features

  1. 01Create or regularize contribution folders
  2. 02Plan claims and supporting evidence
  3. 03Record experiment runs and methods
  4. 04Check data provenance and ethics
  5. 05Build a detailed LaTeX report

Use cases

  1. 01Starting a research contribution
  2. 02Bringing a draft folder into compliance
  3. 03Documenting reproducible experiments
  4. 04Preparing a contribution report

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/vincenzoimp/academic-research-skills --skill develop-contribution

After installing, type/develop-contributionin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

references/
data.md
ethics.md
experiments.md
methodology.md
reproduction.md

Details

From the project README

Develop Contribution

Create or regularize one contributions/<slug>/ folder: badge-general compliant, self-contained, with a report/report.tex detailed enough that paper writing never needs to re-read the code.

Read First

  • contributions/README.md — the badge-general contract
  • survey/survey.pdf, gaps section — when positioning a new contribution
  • references/methodology.md — evaluation, statistics, figure discipline
  • references/experiments.md — run logs and autonomous campaigns
  • references/reproduction.md — reproducing external work
  • references/ethics.md — data red flags
  • references/data.md — data placement and inter-contribution data dependencies (when the contribution has shared/large data or consumes another contribution's output)

Before Building (both modes)

Fill the Positioning section of the contribution README:

  • the claim this contribution will support
  • the delta vs the nearest prior work (cite survey/SOTA citekeys)
  • the evidence plan: what will be measured or collected, and how
  • the falsifiability condition: what outcome would refute the claim

If the claim has no evidence path, stop and discuss with the user.

Create Mode

  1. Copy contributions/_template/ to contributions/<slug>/ (kebab-case slug).
  2. Python code? Write the contribution's own pyproject.toml (name, requires-python, dependencies), add "contributions/<slug>" to [tool.uv.workspace] members in the root pyproject.toml, run uv sync --all-packages from the root (plain uv sync prunes the members' dependencies on a package = false root). Genuinely conflicting dependencies → move the path to exclude, create a local venv, document it in the README.
  3. Develop inside the folder — free-form structure (src/, data/, figures/, outputs/, notebooks). Follow references/methodology.md for any evaluation; keep a run log per references/experiments.md for every run that supports a claim, including autonomous campaigns. For reproductions of external work, follow references/reproduction.md.
  4. Fill the README badge checklist truthfully — every checked box has evidence inside the folder. Check data against references/ethics.md.
  5. Write report/report.tex: motivation, positioning vs the SOTA (cite the root .bib), method, setup, results with figures/tables, limitations. Build it: make contribution C=<slug>.
  6. Run make check.

Regularize Mode

  1. Inventory the draft folder: what exists, what it claims, what is missing against the badge checklist.
  2. Reorganize in place — keep provenance (note original filenames when renaming), capture the environment, verify the run path works end-to-end, record data provenance and ethics flags.
  3. Continue with Create Mode steps 4–6.

Rules

  • Self-contained: no imports from other contributions; tooling worth sharing becomes its own contribution referenced in the README. If the input is another contribution's output, declare it and read it in place — never copy (see references/data.md).
  • Exploratory results never masquerade as confirmed evidence — label them.
  • Negative results are results: report them in report.tex.

Done When

  • Badge checklist fully and truthfully checked, report.pdf built, workspace coherent, make check passes

Comments

Markdown supported