researchforge-paper · git:20260713.896358d · 2026-07-13 · sha256 b67e53bc0ef92dbd
researchforge-paper git:20260713.896358dA
Immutable. This exact content never changes and is served at /api/v1/blob/b67e53bc0ef92dbd.
No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_hbmabtrwag4jkrzh/b67e53bc0ef92dbd7faa460f30099dfda18eec9a32c412cf0cbde9779e5e34f9.json
Then /plugin install researchforge-paper@markdownregistry-hbmabtrwag4jkrzh in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.
--- name: researchforge-paper description: Build the research package — BibTeX citations, related work, evidence matrix, paper outline, reproducibility bundle, and experiment data. Use when the user wants publication materials or a research write-up bundle. --- # Build the research package ```bash researchforge paper package --json ``` Requires stored papers (run the researchforge-papers skill first). Output goes to `.researchforge/research-output/` by default; `--output DIR` writes elsewhere (only with the user's say-so — it creates a directory in their tree). The bundle (all generated from recorded data only): - `citations.bib`, `related_work.md`, `evidence_matrix.csv` — from stored papers and the landscape's evidence claims; - `hypotheses.md`, `paper_outline.md` — the outline marks every section that has no recorded data as *to be completed by the author*; leave those honest, do not fill them with invented content; - `methodology.md`, `limitations.md`, `reproducibility.md` — frozen contract, environment fingerprint, exact command sequence, patch hashes; - `experiments/` — `results.csv`, `run_manifest.json`, and `rejected_experiments.md` (negative results are part of the record); - `figures/README.md` — figures are not auto-generated; say so if asked. When presenting the package, describe what each file contains using the `--json` file list, and remind the user the outline is a starting point for *their* paper: the recorded evidence supports it, but the prose and any novelty claims are theirs to make. ## Rules - The Python engine is the boundary: never work around a validation error, a protected path, or an approval gate — fix the artifact or ask the user. - Approvals belong to the user: never pass `--yes` or type a confirmation unless the user explicitly approved that step in this conversation. - Ground every summary in stored data: quote only numbers returned by `--json` output or files under `.researchforge/` — never invent metrics.