medsci-skills 4.1.0

package/skills/self-review/tests/test_panel_mode.sh

@@ -0,0 +1,69 @@
+#!/usr/bin/env bash
+# Structural + PII regression tests for self-review --panel mode (Phase 2.6).
+# Pure file checks — no agent invocation. Run from anywhere.
+
+set -uo pipefail
+
+REPO_ROOT="$(cd "$(dirname "$0")/../../.." && pwd)"
+SKILL="$REPO_ROOT/skills/self-review/SKILL.md"
+TEMPLATE="$REPO_ROOT/skills/self-review/references/panel_review_template.md"
+PROBE_DIR="$REPO_ROOT/skills/self-review/references/domain-probes"
+
+fail=0
+ran=0
+check() {
+    local label="$1"; shift
+    ran=$((ran + 1))
+    if "$@" >/dev/null 2>&1; then
+        printf '  PASS  %s\n' "$label"
+    else
+        printf '  FAIL  %s\n' "$label"
+        fail=$((fail + 1))
+    fi
+}
+
+[[ -f "$SKILL" ]] || { echo "ENV-ERR: SKILL.md missing" >&2; exit 2; }
+
+# 1. Panel section + flag present.
+check "Phase 2.6 panel section present"       grep -qE '^### Phase 2\.6: Multi-Agent Panel Review' "$SKILL"
+check "--panel flag documented in Optional Flags" grep -qE '^- `--panel`:' "$SKILL"
+check "default stays single-pass (off by default)" grep -qi 'off by default' "$SKILL"
+check "--fix x --panel guard documented"       grep -qi 'Do \*\*not\*\* combine with `--fix`' "$SKILL"
+
+# 2. Reviewer-set mapping names all six domain-probe modules.
+for m in survival_prognostic sr_ma radiomics narrative_review observational_confounding ai_overclaiming; do
+    check "reviewer-set table references $m module" grep -qE "domain-probes/$m\.md" "$SKILL"
+done
+
+# 3. Template exists and carries the schema fields + skeletons.
+check "panel_review_template.md exists"         test -f "$TEMPLATE"
+check "template references skill-dir path in SKILL.md" grep -qE 'references/panel_review_template\.md' "$SKILL"
+check "template has reviewer schema fields"     grep -qE '"severity": "Fatal \| Fixable"' "$TEMPLATE"
+check "template has editor synthesis skeleton"  grep -qi 'Editor synthesis prompt skeleton' "$TEMPLATE"
+
+# 4. Vendored probe modules each reviewer loads exist.
+for m in survival_prognostic sr_ma radiomics narrative_review observational_confounding ai_overclaiming; do
+    check "vendored module present: $m.md"      test -f "$PROBE_DIR/$m.md"
+done
+
+# 5. PoC-content-leak scan on the PoC-derived template.
+#    The authoritative PII gate for names / mentors / institutions / project
+#    codes is scripts/check_precedent.py (SHA-256 hashed, run repo-wide by
+#    validate_skills.sh). Do NOT duplicate those identifiers in cleartext here —
+#    that would re-introduce the very PII as a public string (self-doxxing,
+#    oss-publication-pii-guard §5). This check is belt-and-suspenders for the
+#    non-identifying content markers of the source PoC manuscript (a personal
+#    path, the study disease term, the cohort N) that would signal the template
+#    was not fully generalized. POSIX-safe (no \b).
+POC_MARKERS='/Users/|/home/|emphysema|41,?291'
+ran=$((ran + 1))
+if grep -nEi "$POC_MARKERS" "$TEMPLATE"; then
+    printf '  FAIL  PoC-content-leak scan (panel_review_template.md)\n'
+    fail=$((fail + 1))
+else
+    printf '  PASS  PoC-content-leak scan (panel_review_template.md)\n'
+fi
+
+echo ""
+echo "ran=$ran fail=$fail"
+[[ $fail -eq 0 ]]

package/skills/self-review/tests/test_reference_adequacy.sh

@@ -0,0 +1,68 @@
+#!/usr/bin/env bash
+# Regression test for the reference-adequacy gate (self-review Phase 2.5c-2 /
+# write-paper Step 7.3c). Synthetic fixtures reproduce:
+#   1. an original-research draft whose Statistical Analysis subsection names a
+#      competing-risk model, multiple imputation, the E-value, and an eGFR
+#      equation with ZERO citations (the highest-value failure mode);
+#   2. the same draft with a citation added in each named-method paragraph;
+#   3. a letter whose lower reference target must not false-fail.
+# Stdlib-only (python3).
+set -u
+
+HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+SCRIPT="$HERE/../scripts/check_reference_adequacy.py"
+FIX="$HERE/fixtures"
+OUT="$(mktemp -t ra_XXXX).json"
+trap 'rm -f "$OUT"' EXIT
+
+fail=0
+check() { local label="$1"; shift
+    if "$@" >/dev/null 2>&1; then printf '  PASS  %s\n' "$label"
+    else printf '  FAIL  %s\n' "$label"; fail=$((fail+1)); fi
+}
+# assert a python boolean expression over the JSON at $OUT; `d` is the parsed dict.
+# (No assertion message: the expression embeds single quotes — the check() label
+# already names the case on FAIL.)
+jassert() { python3 -c "
+import json
+d=json.load(open('$OUT'))
+assert ($1)
+"; }
+
+[[ -f "$SCRIPT" ]] || { echo "ENV-ERR: script missing: $SCRIPT" >&2; exit 2; }
+
+echo "== Fixture 1: original research, Methods names methods with 0 citations =="
+python3 "$SCRIPT" --manuscript "$FIX/refadeq_original_uncited.md" \
+    --article-type original_article --out "$OUT" --strict >/dev/null 2>&1
+check "exit 1 under --strict (Major present)"      test "$?" -eq 1
+check "JSON artifact written"                       test -s "$OUT"
+check "methods_zero_citations is true"              jassert "d['methods_zero_citations'] is True"
+check "uncited_named_methods is non-empty"          jassert "len(d['uncited_named_methods']) > 0"
+check "Fine-Gray flagged uncited"                   jassert "'Fine-Gray' in d['uncited_named_methods']"
+check "reference_count_verdict BELOW_TARGET"        jassert "d['reference_count_verdict'] == 'BELOW_TARGET'"
+check "a methods_zero_citations Major finding"      jassert "any(f['subtype']=='methods_zero_citations' and f['severity']=='major' for f in d['findings'])"
+check "all findings fixable_by_ai=false"            jassert "all(f['fixable_by_ai'] is False for f in d['findings'])"
+
+echo "== Fixture 2: same draft, each named method now cited =="
+python3 "$SCRIPT" --manuscript "$FIX/refadeq_original_fixed.md" \
+    --article-type original_article --out "$OUT" --strict >/dev/null 2>&1
+check "exit 0 (no Major; Methods gap cleared)"      test "$?" -eq 0
+check "methods_zero_citations is false"             jassert "d['methods_zero_citations'] is False"
+check "no methods_named_method_uncited finding"     jassert "not any(f['subtype']=='methods_named_method_uncited' for f in d['findings'])"
+check "adequacy_safe is true"                        jassert "d['adequacy_safe'] is True"
+
+echo "== Fixture 3: letter (lower target must not false-fail) =="
+python3 "$SCRIPT" --manuscript "$FIX/refadeq_letter.md" \
+    --article-type letter --out "$OUT" --strict >/dev/null 2>&1
+check "exit 0 (letter target met)"                  test "$?" -eq 0
+check "verdict ADEQUATE for letter"                 jassert "d['reference_count_verdict'] == 'ADEQUATE'"
+check "no adequacy findings"                         jassert "len(d['findings']) == 0"
+
+echo "== Alias map: nhis_cohort routes to original_research target =="
+python3 "$SCRIPT" --manuscript "$FIX/refadeq_original_uncited.md" \
+    --article-type nhis_cohort --out "$OUT" --quiet >/dev/null 2>&1
+check "nhis_cohort -> original_research bucket"      jassert "d['article_bucket'] == 'original_research'"
+check "original_research target is [25,45]"         jassert "d['effective_target'] == [25,45]"
+
+echo "fail=$fail"; [[ "$fail" -eq 0 ]] && echo "ALL PASS" || echo "FAILURES: $fail"
+exit "$fail"

package/skills/self-review/tests/test_reviewer_team_consistency.sh

@@ -0,0 +1,138 @@
+#!/usr/bin/env bash
+# Regression tests for self-review check_reviewer_team_consistency.py.
+
+set -uo pipefail
+
+REPO_ROOT="$(cd "$(dirname "$0")/../../.." && pwd)"
+SCRIPT="$REPO_ROOT/skills/self-review/scripts/check_reviewer_team_consistency.py"
+TMP="$(mktemp -d -t reviewer_team.XXXXXX)"
+trap 'rm -rf "$TMP"' EXIT
+
+[[ -f "$SCRIPT" ]] || { echo "ENV-ERR: script missing" >&2; exit 2; }
+
+fail=0
+ran=0
+assert_exit() {
+    local label="$1" expected="$2" actual="$3"
+    ran=$((ran + 1))
+    if [[ "$expected" == "$actual" ]]; then
+        printf '  PASS  %-50s exit=%s\n' "$label" "$actual"
+    else
+        printf '  FAIL  %-50s expected=%s actual=%s\n' "$label" "$expected" "$actual"
+        fail=$((fail + 1))
+    fi
+}
+
+# --------------------------------------------------------------------------
+# Case 1: clean — dual review claimed, no limits confession.
+# --------------------------------------------------------------------------
+cat > "$TMP/c1.md" <<'EOF'
+## **METHODS**
+Two reviewers independently screened titles and abstracts.
+
+## **DISCUSSION**
+The cohort overlap reduces effective sample size.
+EOF
+python3 "$SCRIPT" --manuscript "$TMP/c1.md" \
+    --out "$TMP/c1.md.audit" --quiet
+assert_exit "case 1: dual claim, no limits confession" 0 $?
+
+# --------------------------------------------------------------------------
+# Case 2: fabrication-grade — Methods dual + Limitations single confession.
+# --------------------------------------------------------------------------
+cat > "$TMP/c2.md" <<'EOF'
+## **METHODS**
+Two reviewers independently screened titles and abstracts and extracted data.
+
+## **DISCUSSION**
+
+### Limitations
+We used a single primary reviewer for data extraction; a 20% sample by an
+additional reviewer is deferred to before submission.
+EOF
+python3 "$SCRIPT" --manuscript "$TMP/c2.md" \
+    --out "$TMP/c2.md.audit" --quiet
+assert_exit "case 2: dual + single confession (FAIL)" 1 $?
+
+# Verify the markdown report flags both claim sides.
+grep -q "MAJOR red flag" "$TMP/c2.md.audit" || { echo "  FAIL c2 markdown body"; fail=$((fail + 1)); }
+grep -q "single primary reviewer" "$TMP/c2.md.audit" || { echo "  FAIL c2 markdown body"; fail=$((fail + 1)); }
+
+# --------------------------------------------------------------------------
+# Case 3: PROSPERO dual + Limitations single. Also FAIL.
+# --------------------------------------------------------------------------
+cat > "$TMP/c3.md" <<'EOF'
+## **METHODS**
+Records were screened against pre-specified eligibility criteria.
+
+## **DISCUSSION**
+### Limitations
+A single primary reviewer extracted data due to resource constraints.
+EOF
+cat > "$TMP/c3_prospero.md" <<'EOF'
+# PROSPERO record
+Two independent reviewers will perform full-text screening and data extraction.
+EOF
+python3 "$SCRIPT" --manuscript "$TMP/c3.md" --prospero "$TMP/c3_prospero.md" \
+    --out "$TMP/c3.md.audit" --quiet
+assert_exit "case 3: PROSPERO dual + limits single (FAIL)" 1 $?
+
+# --------------------------------------------------------------------------
+# Case 4: single confession alone (no dual claim) => PASS.
+# --------------------------------------------------------------------------
+cat > "$TMP/c4.md" <<'EOF'
+## **METHODS**
+Data extraction was performed by the first reviewer.
+
+## **DISCUSSION**
+### Limitations
+A single primary reviewer extracted data; this is a limitation of our review.
+EOF
+python3 "$SCRIPT" --manuscript "$TMP/c4.md" \
+    --out "$TMP/c4.md.audit" --quiet
+assert_exit "case 4: single confession only (PASS)" 0 $?
+
+# --------------------------------------------------------------------------
+# Case 5: extraction JSON names an LLM as a reviewer => fatal FAIL.
+# --------------------------------------------------------------------------
+cat > "$TMP/c5.md" <<'EOF'
+## **METHODS**
+Records were screened against pre-specified eligibility criteria.
+EOF
+cat > "$TMP/c5_extract.json" <<'EOF'
+{"study": "Example 2024", "reviewer_1": "Claude", "reviewer_2": "Jane Doe"}
+EOF
+python3 "$SCRIPT" --manuscript "$TMP/c5.md" --extraction-json "$TMP/c5_extract.json" \
+    --out "$TMP/c5.md.audit" --quiet
+assert_exit "case 5: LLM-as-reviewer in extraction JSON (FAIL)" 1 $?
+grep -q "LLM-AS-REVIEWER" "$TMP/c5.md.audit" || { echo "  FAIL c5 markdown body"; fail=$((fail + 1)); }
+
+# --------------------------------------------------------------------------
+# Case 6: future-tense deferred mitigation => MAJOR FAIL.
+# --------------------------------------------------------------------------
+cat > "$TMP/c6.md" <<'EOF'
+## **METHODS**
+Records were screened by the primary reviewer.
+
+## **DISCUSSION**
+### Limitations
+A duplicate-screening check will be completed before submission.
+EOF
+python3 "$SCRIPT" --manuscript "$TMP/c6.md" \
+    --out "$TMP/c6.md.audit" --quiet
+assert_exit "case 6: deferred mitigation (FAIL)" 1 $?
+grep -q "DEFERRED-MITIGATION" "$TMP/c6.md.audit" || { echo "  FAIL c6 markdown body"; fail=$((fail + 1)); }
+
+# --------------------------------------------------------------------------
+# Case 7: clean extraction JSON (human reviewers) => PASS (backward compat).
+# --------------------------------------------------------------------------
+cat > "$TMP/c7_extract.json" <<'EOF'
+{"study": "Example 2024", "reviewer_1": "Jane Doe", "reviewer_2": "John Roe"}
+EOF
+python3 "$SCRIPT" --manuscript "$TMP/c1.md" --extraction-json "$TMP/c7_extract.json" \
+    --out "$TMP/c7.md.audit" --quiet
+assert_exit "case 7: human reviewers in extraction JSON (PASS)" 0 $?
+
+echo ""
+echo "ran=$ran fail=$fail"
+[[ $fail -eq 0 ]]

package/skills/self-review/tests/test_scope_coherence.sh

@@ -0,0 +1,46 @@
+#!/usr/bin/env bash
+# Regression test for the scope-coherence gate (self-review §D).
+# Synthetic, PII-free fixtures reproduce: (a) a cross-sectional design with a
+# prognostic/surveillance conclusion (CROSS_SECTIONAL_PROGNOSTIC), (b) a binary
+# surrogate endpoint driving a care directive (SURROGATE_CARE_DIRECTIVE). The clean
+# fixture is a longitudinal cohort with an association conclusion.
+# Stdlib-only (python3).
+set -u
+
+HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+SCRIPT="$HERE/../scripts/check_scope_coherence.py"
+BAD="$HERE/fixtures/scope_bad.md"
+SURR="$HERE/fixtures/scope_surrogate.md"
+CLEAN="$HERE/fixtures/scope_clean.md"
+OUT="$(mktemp -t scope_XXXX).json"
+trap 'rm -f "$OUT"' EXIT
+
+fail=0
+check() { local label="$1"; shift
+    if "$@" >/dev/null 2>&1; then printf '  PASS  %s\n' "$label"
+    else printf '  FAIL  %s\n' "$label"; fail=$((fail+1)); fi
+}
+has_verdict() { python3 -c "
+import json,sys
+d=json.load(open('$OUT'))
+assert any(c['verdict']=='$1' for c in d['claims']), '$1 not found'
+"; }
+
+[[ -f "$SCRIPT" ]] || { echo "ENV-ERR: script missing" >&2; exit 2; }
+
+# (1) cross-sectional + prognostic conclusion -> CROSS_SECTIONAL_PROGNOSTIC, exit 1
+python3 "$SCRIPT" --manuscript "$BAD" --out "$OUT" --strict --quiet >/dev/null 2>&1
+check "exit 1 (cross-sectional prognostic)" test "$?" -eq 1
+check "CROSS_SECTIONAL_PROGNOSTIC detected" has_verdict CROSS_SECTIONAL_PROGNOSTIC
+
+# (2) binary surrogate + care directive -> SURROGATE_CARE_DIRECTIVE, exit 1
+python3 "$SCRIPT" --manuscript "$SURR" --out "$OUT" --strict --quiet >/dev/null 2>&1
+check "exit 1 (surrogate care directive)" test "$?" -eq 1
+check "SURROGATE_CARE_DIRECTIVE detected" has_verdict SURROGATE_CARE_DIRECTIVE
+
+# (3) longitudinal cohort + association conclusion -> exit 0
+python3 "$SCRIPT" --manuscript "$CLEAN" --strict --quiet >/dev/null 2>&1
+check "exit 0 on clean manuscript" test "$?" -eq 0
+
+echo "fail=$fail"; [[ "$fail" -eq 0 ]] && echo "ALL PASS" || echo "FAILURES: $fail"
+exit "$fail"

package/skills/setup-medsci/SKILL.md

@@ -0,0 +1,110 @@
+---
+name: setup-medsci
+description: Diagnostic checklist for the MedSci Skills runtime. Verifies Python, R, Node, Claude Code, Git, Zotero, and configured MCP servers, and prints a pass/fail table with links to the right setup doc for any missing component. Read-only — does not install anything.
+triggers: setup, install, environment, diagnostic, check setup, why doesn't this work, missing python, missing R, MCP not connected, 환경 설정, 설치 점검
+tools: Bash, Read
+model: inherit
+---
+
+# Setup-MedSci Skill
+
+You are helping a medical researcher verify that their environment is correctly configured to run MedSci Skills. You **do not install anything** — you only diagnose what's present, what's missing, and where to find the setup doc for any missing component. This skill is intentionally read-only so that a doctor can run it safely without worrying about breaking their system.
+
+## Communication Rules
+
+- Communicate with the user in their preferred language (Korean or English).
+- The diagnostic table itself is in English so it can be pasted into GitHub issues for support.
+- All recommended remediation links point to `docs/setup/` markdown guides in the medsci-skills repo.
+
+## Workflow
+
+### Phase 1: Detect OS
+
+Run:
+```bash
+uname -s
+```
+
+- `Darwin` → macOS path → recommend `docs/setup/mac.md`
+- `Linux` → Linux path (uses similar tooling to Mac) → recommend `docs/setup/mac.md`
+- `MINGW*`, `MSYS*`, `CYGWIN*`, or detection failure on Windows → recommend `docs/setup/windows.md`
+
+### Phase 2: Run Diagnostic Commands
+
+For each tool, run the command and capture both the version output and the exit code:
+
+| Tool | Command | Required version |
+|---|---|---|
+| Python | `python3 --version` (Mac/Linux) or `python --version` (Windows) | 3.11 or higher |
+| R | `Rscript --version` (writes to stderr) | 4.0 or higher |
+| Node.js | `node --version` | v20 or higher |
+| Git | `git --version` | 2.30 or higher |
+| Claude Code CLI | `claude --version` | any |
+| MCP servers | `claude mcp list` | at least one of: zotero, gdrive, pubmed, filesystem |
+
+Use `command -v <tool>` first to detect presence without running it (avoids triggering long initialization).
+
+### Phase 3: Emit Checklist Table
+
+Print a single Markdown table to stdout in this exact format:
+
+```
+## MedSci Skills Setup Diagnostic
+
+OS detected: <macOS | Linux | Windows>
+Date: <YYYY-MM-DD>
+
+| Component | Status | Detected | Required | Action |
+|---|:---:|---|---|---|
+| Python 3.11+ | ✅ / ❌ | 3.11.9 | 3.11+ | OK / See docs/setup/mac.md Step 2 |
+| R 4.x | ✅ / ❌ | 4.3.1 | 4.0+ | OK / See docs/setup/mac.md Step 3 |
+| Node.js 20+ | ✅ / ❌ | v20.11.1 | v20+ | OK / See docs/setup/mac.md Step 4 |
+| Git | ✅ / ❌ | 2.42.0 | 2.30+ | OK / See docs/setup/<os>.md Step 5 |
+| Claude Code CLI | ✅ / ❌ | 1.5.x | any | OK / See docs/setup/<os>.md Step 6 |
+| MCP: zotero | ✅ / ❌ / ⚠️ | Connected | optional | OK / See docs/setup/mcp-setup.md |
+| MCP: gdrive | ✅ / ❌ / ⚠️ | Connected | optional | OK / See docs/setup/mcp-setup.md |
+| MCP: filesystem | ✅ / ❌ | Connected | recommended | OK / See docs/setup/mcp-setup.md |
+
+Summary: <X required components passed, Y missing>
+Next step: <one-sentence action>
+```
+
+Status legend:
+- ✅ Present and meets minimum version
+- ❌ Missing or below minimum version
+- ⚠️ Present but optional and not connected
+
+### Phase 4: Suggest Remediation
+
+If everything ✅: "Your environment is ready. Try Demo 1 with `cd ~/medsci-skills/demo/01_wisconsin_bc && claude '/orchestrate --e2e'`."
+
+If anything ❌ in **required** rows: print the corresponding doc link. Do **not** offer to install — direct the user to follow the doc step. The doc tells them exactly what to copy-paste.
+
+If only **optional MCP** rows are ❌: explain that MedSci Skills work without MCP servers but `lit-sync`, `verify-refs`, and `write-paper` are smoother with Zotero MCP. Offer the `docs/setup/mcp-setup.md` link.
+
+## Reference Files
+
+- `references/setup-checklist.md` — verbatim list of every check this skill runs and the corresponding documentation link
+
+## Output Contract
+
+| Artifact | Filename | Format |
+|----------|----------|--------|
+| Diagnostic report | stdout (Markdown table) | Markdown |
+| Optional log | `~/.medsci-skills/diagnostic-YYYY-MM-DD.md` | Markdown |
+
+If the user asks for a copyable report (e.g., for a GitHub issue), write the diagnostic to the optional log path and tell them where it is.
+
+## What This Skill Does NOT Do
+
+- **Does not install anything.** No `brew install`, `winget install`, `pip install`, `Rscript -e 'install.packages(...)'`, or any other state-changing command. Read-only diagnostics only.
+- **Does not modify `~/.claude.json` or any MCP configuration.** It only reads `claude mcp list` output.
+- **Does not check skill versions or skill content.** That is the job of `validate_skills.sh` and `manage-project status`.
+- **Does not auto-fix anything.** If a tool is missing, the user must go to the setup doc and follow the steps themselves. This is intentional — auto-installers for system Python and R are a support nightmare for non-developer users.
+
+## Anti-Hallucination
+
+- **Never fabricate version numbers.** Always run the actual command and report the exact stdout. If the command fails, report the failure verbatim — do not guess what version is "probably" installed.
+- **Never invent doc paths.** The five setup docs are: `docs/setup/README.md`, `docs/setup/mac.md`, `docs/setup/windows.md`, `docs/setup/mcp-setup.md`, `docs/setup/common-issues.md`. Do not link to a doc that does not exist in the repo.
+- **Never claim an MCP server works without verifying via `claude mcp list`.** A configured-but-disconnected server is ⚠️, not ✅.
+- **If `command -v` reports a tool present but the version flag fails**, mark the row ❌ and report the failure command — the install is broken, not just outdated.

package/skills/setup-medsci/references/setup-checklist.md

@@ -0,0 +1,51 @@
+# Setup Diagnostic Checklist
+
+Verbatim list of every check `setup-medsci` runs, with the exact command, the parse rule, and the documentation link to print on failure.
+
+## Required Components
+
+| ID | Component | Detect command | Version command | Min version | Doc link on failure |
+|---|---|---|---|---|---|
+| R1 | Python 3 | `command -v python3` | `python3 --version` | 3.11.0 | `docs/setup/mac.md#step-2` (Mac) / `docs/setup/windows.md#step-3` (Windows) |
+| R2 | R | `command -v Rscript` | `Rscript --version` (stderr) | 4.0.0 | `docs/setup/mac.md#step-3` / `docs/setup/windows.md#step-4` |
+| R3 | Node.js | `command -v node` | `node --version` | 20.0.0 | `docs/setup/mac.md#step-4` / `docs/setup/windows.md#step-5` |
+| R4 | Git | `command -v git` | `git --version` | 2.30.0 | `docs/setup/mac.md#step-5` / `docs/setup/windows.md#step-6` |
+| R5 | Claude Code CLI | `command -v claude` | `claude --version` | any (latest recommended) | `docs/setup/mac.md#step-6` / `docs/setup/windows.md#step-7` |
+
+## Optional Components
+
+| ID | Component | Detect command | Status rule | Doc link on missing |
+|---|---|---|---|---|
+| O1 | MCP: zotero | `claude mcp list` | ✅ if "zotero ✓ Connected", ⚠️ if listed but not connected, ❌ if absent | `docs/setup/mcp-setup.md#zotero-mcp` |
+| O2 | MCP: gdrive | `claude mcp list` | ✅ if "gdrive ✓ Connected", ⚠️ if listed but not connected, ❌ if absent | `docs/setup/mcp-setup.md#google-drive--workspace-mcp` |
+| O3 | MCP: pubmed | `claude mcp list` | ✅ if "pubmed ✓ Connected", ⚠️ if listed but not connected, ❌ if absent | `docs/setup/mcp-setup.md#pubmed-mcp-alternative-to-built-in-search-lit` |
+| O4 | MCP: filesystem | `claude mcp list` | ✅ if "filesystem ✓ Connected", ⚠️ if listed but not connected, ❌ if absent | `docs/setup/mcp-setup.md#filesystem-mcp-built-in-usually-pre-configured` |
+| O5 | Zotero desktop | `pgrep -x Zotero` (Mac) / `tasklist /FI "IMAGENAME eq zotero.exe"` (Win) | ✅ if running, ⚠️ if installed but not running, ❌ if not installed | `docs/setup/mac.md#step-7` / `docs/setup/windows.md#step-8` |
+
+## Version Comparison Rule
+
+For required-version checks (R1-R4), parse the major.minor version from stdout, compare against minimum. If parse fails, mark ❌ and report the parse failure verbatim.
+
+Examples of parsing:
+- `Python 3.11.9` → `3.11`, compare to `3.11` → pass
+- `Python 3.10.13` → `3.10`, compare to `3.11` → fail
+- `R scripting front-end version 4.3.1 (2023-06-16)` → `4.3`, compare to `4.0` → pass
+- `v20.11.1` → `20.11`, compare to `20.0` → pass
+- `git version 2.42.0` → `2.42`, compare to `2.30` → pass
+
+## Summary Rules
+
+After running all checks:
+- **All required ✅**: print "Your environment is ready. Try Demo 1 with `cd ~/medsci-skills/demo/01_wisconsin_bc && claude '/orchestrate --e2e'`."
+- **One or more required ❌**: print "Setup incomplete. Address the ❌ rows above using the linked docs, then re-run `/setup-medsci`."
+- **All required ✅, optional MCP all ❌**: print "Core environment ready. Optional MCP servers (Zotero / Google Drive / PubMed) are not configured — see `docs/setup/mcp-setup.md` for richer integration."
+
+## What This Skill Does NOT Check
+
+To keep the diagnostic fast and read-only, these are explicitly out of scope:
+- Python package installs (matplotlib, pandas, scikit-learn, etc.) — checked by individual analysis skills when invoked
+- R package installs (metafor, survey, etc.) — checked by `analyze-stats` and `meta-analysis` when invoked
+- Disk space — irrelevant for skill execution
+- Anthropic API quota / rate limits — visible only at runtime
+- Claude Code skill installation — `validate_skills.sh` covers this
+- Zotero library content / collection structure — `lit-sync` and `verify-refs` cover this

package/skills/setup-medsci/skill.yml

@@ -0,0 +1,30 @@
+schema_version: 2
+name: setup-medsci
+layer: A
+owner_domain: environment_setup
+
+when_to_use: "Set up the local environment for MedSci Skills (dependencies, tooling checks)."
+when_NOT_to_use: "Installing the skills themselves (use installers/install.py)."
+
+inputs:
+  - "local machine state"
+outputs:
+  - "verified environment / dependency report"
+side_effects:
+  - runs_setup_commands
+downstream_consumers:
+  - none
+forbidden_actions:
+  - install_without_user_consent_for_destructive_actions
+
+# v2.1 quality card
+purpose: "Check and prepare the local toolchain (Python/R/CLI deps) needed to run the skills."
+safety_boundaries:
+  - "Confirms before destructive or system-wide changes; surfaces what is missing."
+  - "Operates locally; does not transmit machine state."
+known_limitations:
+  - "Environment coverage is best-effort across OSes; some hosts need manual steps."
+  - "No standalone demo; an operational utility."
+validation_commands:
+  - "re-run the setup check and confirm all dependencies report present"
+evidence_surface: manual_workflow