feat(v3): Tier 0 polish — CITATION.cff, Zenodo workflow, clinician setup onboarding

[Yoojin-nam]

Summary

Internal polish before external promotion to awesome-claude-code (re-submission, prior #1389/#1518), OpenClaw aggregator, and Anthropic DevRel. Addresses four blockers identified in the 2026-05-13 audit vs K-Dense scientific-agent-skills (20.8k stars) and OpenClaw Medical Skills (2.5k stars).

What changed

• Academic backlink — `CITATION.cff` (cff-version 1.2.0) + `.zenodo.json`. DOI populates after first Zenodo archive of a tagged release.
• Release automation — `.github/workflows/release.yml` on `v*` tag push: builds classroom ZIPs, creates GitHub Release with notes from CHANGELOG, attaches ZIPs. Zenodo integration toggled once at zenodo.org/account/settings/github.
• Clinician onboarding — five-doc guide at `docs/setup/` (`README.md`, `mac.md`, `windows.md`, `mcp-setup.md`, `common-issues.md`) + new `setup-medsci` skill that runs `which / claude mcp list` diagnostics and prints a checklist with status (✅ / ⚠️ / ❌) and a doc link per missing component. Read-only — does not install anything (intentional: auto-installers for system Python and R are a support nightmare for non-developers).
• Scope clarity — README "What This Is NOT" section explicitly defers broad scientific tooling to K-Dense and biomedical aggregation to OpenClaw, positioning medsci-skills as the opinionated medical manuscript pipeline.
• README — new `Setup` section linking the new docs and `/setup-medsci`, citation badge, DOI badge placeholder.
• CHANGELOG.md — Unreleased entry summarizing all of the above.
• GitHub topics — swapped 4 generic (`ai-tools`, `academic-writing`, `open-source`, `research-tools`) for 4 specific (`agent-skills`, `tripod-ai`, `irb-protocol`, `physician-researcher`) within the 20-topic cap.

Why now

Doctor users coming via Show HN / awesome-claude-code re-submission need to (a) verify the project is academically citable and (b) install Python/R/Claude Code without bouncing. The `docs/setup/` guide and `setup-medsci` skill close the largest onboarding hole identified in user feedback this cycle.

Test plan

• `scripts/validate_skills.sh` PASS (verified locally via pre-commit hook — `setup-medsci` PASSes all required checks; size 110 lines is THIN tier WARN, not FAIL)
• After merge: tag `v3.0.0` → `.github/workflows/release.yml` triggers → GitHub Release with classroom ZIPs created
• Toggle Zenodo GitHub integration → first archive lands → DOI back-filled into README badge + CITATION.cff in a follow-up commit on `main`
• `/setup-medsci` runs end-to-end on a clean machine (will be tested by next external user; instrumented via the diagnostic output)

🤖 Generated with Claude Code