RForge Plugin
R package ecosystem orchestrator for Claude Code — 41 commands, R-aware hooks, validation skills.
Self-contained R package analysis for Claude Code. As of v1.3.0 the plugin is fully self-sufficient — pure-Python lib/ modules handle discovery, dependencies, status, and init. No MCP server required.
What's new in v2.7.0
- 🌌
r:submit --universe — new opt-in flag that adds an R-universe early-access tier. R-universe rebuilds your package from its GitHub repo within minutes and serves CRAN-like binaries, so users can install the new version (install.packages("<pkg>", repos = "https://<owner>.r-universe.dev")) while CRAN's slower human review runs in parallel. The flag auto-detects your universe from the git origin remote (--universe-name <owner> to override), reads the public R-universe API, and reports per-platform build status. It's read-only — R-universe builds on git push, so it never uploads — and the status is advisory in the CRAN checklist: it never blocks the (still manual, never-automatic) CRAN handoff. Backed by new pure-stdlib lib/runiverse.py (urllib-only; no gh/R); degrades to a warn envelope offline/unregistered with one-time setup guidance. Commands unchanged at 35 (a flag, not a new command).
What's new in v2.6.0
- 🚀
r:submit — new per-package command that wraps the moment of CRAN submission. Gates on r:cran-prep being ready, builds the tarball, and cuts a GitHub pre-release (not "Latest") of it with cran-comments.md attached, then prints the CRAN submit checklist — it never auto-submits. r:submit --promote flips the pre-release to a full release once CRAN accepts (gh release edit --prerelease=false --latest). Using a pre-release promoted in place sidesteps the r-pkgs anti-pattern of tagging a final release before acceptance (resubmissions bump the version). gh is a soft dependency with a printed manual-recipe fallback. Backed by pure-Python lib/ghrelease.py. Commands 34 → 35.
What's new in v2.5.0
- 🔗
r:deps-sync — new pure-Python per-package command that reconciles DESCRIPTION against what the code actually uses. Scans R//tests//vignettes/ + NAMESPACE for namespace usage and reports missing (used, undeclared → Imports), misclassified (in Suggests but used unconditionally in R/ → Imports), missing_suggests (tests/vignettes-only), and unused dependencies, plus a suggested DESCRIPTION patch. Report-only by default; --write applies the unambiguous changes. The misclassified finding is the static sibling of r:check --strict's noSuggests pass — it catches the medfit/MASS class before R runs. Intra-package, complementing /rforge:deps (the inter-package graph). Commands 33 → 34.
What's new in v2.4.0
- 🧭 Ecosystem-manifest discovery (
/rforge:detect, /rforge:status) — discovery can now read an optional ecosystem manifest (a curated YAML listing packages with role/repo/cran/status_file), located via a new manifest: key in the root .rforge.yaml. Discovered packages are enriched with that metadata (matched by name, case-insensitive), and any mismatch between the manifest and what's on disk surfaces as drift (manifest_only / disk_only). /rforge:detect shows a manifest: header + per-package role; /rforge:status adds a conditional Role column. Parsed by a vendored YAML-subset reader — discovery.py stays stdlib-only (no PyYAML). Zero behavior change when no manifest is configured. New public API: Manifest, ManifestEntry, Drift, parse_manifest, read_manifest.
What's new in v2.3.0
