orchestrate

Orchestrate Skill

Route research requests in the medsci-skills bundle to the right skill, or chain several skills in the correct order. You classify, plan and delegate: never produce a routed skill's output yourself — invoke the skill that owns it.

---

Available Skills

Skill	Domain	When to Route
search-lit	Literature	Find papers, verify citations, build reference lists, check if a topic has been studied
design-study	Methodology	Review study design, identify leakage/bias, pick reporting guideline, validate analysis plan
intake-project	Project setup	New or messy project folder, "what is this project?", classify and scaffold
manage-project	Project mgmt	Scaffold directories, track progress, generate checklists and timelines
analyze-stats	Statistics	Generate R/Python code for diagnostic accuracy, demographics, meta-analysis stats, agreement, regression (logistic/linear), propensity score, repeated measures
make-figures	Visualization	ROC curves, forest plots, flow diagrams (PRISMA/CONSORT/STARD), Kaplan-Meier, Bland-Altman, visual/graphical abstracts
meta-analysis	Systematic review	Full MA pipeline: protocol, search, screening, extraction, synthesis, PRISMA-DTA
write-paper	Writing	IMRAD manuscript drafting (8-phase pipeline), any section writing
self-review	Quality	Pre-submission self-check with domain probes (Survival / SR-MA / Radiomics / Narrative); optional --panel for a high-stakes final QC pass
check-reporting	Compliance	Audit against 49 reporting guidelines and risk-of-bias tools
revise	Revision	Parse reviewer comments, generate point-by-point response, track changes
grant-builder	Funding	Structure grant proposals: significance, innovation, approach, milestones
present-paper	Presentation	Prepare academic talks: analyze paper, draft scripts, inject slide notes, Q&A prep
publish-skill	Packaging	Convert a personal skill into an open-source distributable package
calc-sample-size	Statistics	Sample size calculation (17 tests including Cox EPV), power analysis, IRB justification text
find-journal	Submission	Journal recommendation based on abstract/scope matching, post-rejection re-targeting
add-journal	Journal DB	Add a new journal to the profile database; extracts metadata from author guidelines
fulltext-retrieval	Literature	Batch download open-access PDFs by DOI using Unpaywall, PMC, OpenAlex APIs
deidentify	Data safety	De-identify clinical data containing PHI before any LLM processing. Standalone Python CLI (no LLM).
clean-data	Data	Data profiling, missing value flagging, outlier detection, cleaning code generation
generate-codebook	Data	Generate a citable data dictionary/codebook from a dataset; flags coded variables as [NEEDS DICTIONARY]; feeds /define-variables
version-dataset	Data	Content-hash manifest of a dataset; verify drift (schema/rows/values) and diff versions; reproducibility lock
write-protocol	Protocol	IRB/ethics protocol drafting, 4 core sections + 6 skeleton sections with TODO markers
define-variables	Operationalization	Literature-grounded variable definitions, cutoffs, DB-variable mappings; runs between /search-lit and /write-protocol for observational studies
verify-refs	Reference audit	Read-only PubMed/CrossRef audit of manuscript references; first-author cross-check; sole writer of qc/reference_audit.json. Never modifies refs
manage-refs	Reference lifecycle	Citekey validation, journal-CSL pandoc rendering, manuscript ↔ DOCX cross-reference QC, [N] ↔ [@key] marker conversion, Zotero CWYW field-code injection. Sole writer of manuscript_final.docx, qc/xref_audit.json
lit-sync	Reference sync	Zotero collection ↔ Better BibTeX manuscript/_src/refs.bib ↔ Obsidian literature notes. Sole writer of refs.bib (auto-export); upstream of manage-refs
obsidian-paper-vault	Vault build	A folder of PDFs → templated Obsidian literature notes + atomic concept notes synthesized across them. Enters the same vault folders as lit-sync from the PDF side
humanize	Quality	AI-pattern density sweep (<2.0/1000 words target); rewrites flagged passages while preserving technical accuracy. Phase 7.5 of write-paper
academic-aio	Visibility	AI-search-engine optimization for medical AI papers (Perplexity, ChatGPT web, Elicit, Consensus, SciSpace, RAG tools). Opt-in checklist; never auto-applies edits
render-pdf-doc	Document layout	Non-bibliography academic markdown → PDF (proposal, briefing, anchor doc, IRB cover, reference table), CJK-aware. Boundary opposite of manage-refs scripts/render_pandoc.sh
fill-protocol	Form filling	Institutional Word form filling (.doc/.docx) for IRB/ethics/grant templates; renders write-protocol content into the institutional template
fill-icmje-coi	Form filling	Batch ICMJE COI Disclosure Form generation per author from a synthetic seed
sync-submission	Submission	SSOT-to-submission drift audit; journal-specific submission manifest creation from canonical manuscript artifacts
peer-review	Review	External manuscript peer review draft generation (journal-specific formatting). Use ONLY for reviewing other authors' work, never for self-review
review-paper	Writing	Scaffold/draft a literature review (narrative / scoping PRISMA-ScR / systematic). Distinct from write-paper (original research) and meta-analysis (pooling)
polish-language	Quality	Academic-English consistency lint + non-native clarity polish (abbreviations, US/UK spelling, ranges, P/p case, units). Style-only; distinct from humanize (AI-tell removal) and check-reporting (guideline items)
author-strategy	Analysis	PubMed author-profile analysis: study-type classification, trajectory-archetype, publication-strategy report from a name
batch-cohort	Analysis	Generate N analysis scripts from one validated methodology template × many exposure/outcome combinations (same method, swap variables) + summary matrix
replicate-study	Analysis	Replicate an existing cohort study's methodology on a different database: design extraction, variable-harmonization table, replication-difference report
cross-national	Analysis	Cross-national comparison study (KNHANES + NHANES + CHNS or parallel surveys): variable harmonization + parallel weighted analysis
ma-scout	Systematic review	Meta-analysis topic discovery + feasibility (professor-first profile→gap, or topic-first question→gap→co-author) before a protocol exists
find-cohort-gap	Methodology	Research-gap discovery from a longitudinal cohort DB: profile strengths, match PI expertise, literature-saturation scan, ranked topic proposals
design-ai-benchmarking	Methodology	Design/validity review for benchmarking AI systems against a human-expert reference panel (rubrics, calibration probes, panel construction, IRR targets) — before data collection
model-selection	Modeling	Choose the model for an imaging study: a paper-grounded architecture family (task + modality + data scale + imbalance → shortlist), then vet the concrete repo / checkpoint (licence, version pin, weight provenance; flags an evaluation arm on the benchmark the model was developed or tuned on)
imaging-data	Modeling	Profile an imaging dataset before any modelling decision (spacing/orientation, intensity domain, label integrity, foreground fraction, target volume) and gate it against the plan; then plan/audit DICOM/NIfTI intake, resampling, normalisation and augmentation with a preprocessing manifest + data-stage leakage gate before model-scaffold
model-scaffold	Modeling	Generate a reproducible runnable PyTorch training repo (patient-level seed-locked split, task model, train/eval scripts, repro record) — between choosing an architecture and validating a trained model
radiomics-ml	Modeling	Produce/audit a radiomics / tabular-ML study (imaging or clinical features → penalised logistic / SVM / RF / gradient-boosting / MLP → outcome) with a nested-CV / feature-stability / calibration / external-validation gate (no GPU)
model-assessment	Validation	Validate/evaluate a trained imaging model: split-leakage gate + validation design (internal vs external, comparator, sizing); task-correct held-out metrics (Dice + boundary, AUROC + AUPRC with bootstrap CIs, FROC/mAP, calibration, subgroups) → per-case table; uncertainty/OOD/abstention for deployment claims; Grad-CAM/explainability sanity checks + localisation
mllm-eval	Evaluation	Design/audit an evaluation harness for an LLM/MLLM clinical task (report generation, VQA, extraction/classification): adjudicated reference, clinical-efficacy metrics beyond BLEU/ROUGE, hallucination, contamination, prompt-sensitivity, reader study
model-card	Documentation	Generate a Model Card + Datasheet + data-quality pass for an engineer-built imaging model from user-supplied facts, with a completeness gate
contribute	Setup	Offer a local edit (a journal profile, a fix) back as a pull request or issue without typing git; blocks on patient data in the diff and sends nothing until confirmed. Also files a false positive or a failed step
setup-medsci	Setup	Read-only runtime diagnostic (Python, R, Node, Claude Code, Git, Zotero, MCP servers): pass/fail table with the setup doc for each missing component

---

Classification Logic

Classify each request into one of these intents.

Single-skill requests (route directly)

Route to the skill whose Available Skills row matches: name it with a one-line reason, then invoke it. The rows below cover modes and flags, look-alike skills, and non-English phrasings:

User says something like...	Route to
"I have a messy folder, help me organize"	/intake-project
"Set up a new project" / "Create project scaffold"	/manage-project init
"Review my manuscript before submission"	/self-review
"Brutally / harshly check before submission" / "top-tier journal final check" / "multi-reviewer / panel review" / "review it from stats, clinical, and imaging angles" / "혹독하게 제출 전 점검"	/self-review --panel --json
"Review someone else's manuscript" / "Journal club critique draft" / "외부 논문 리뷰 답변"	/peer-review
"Which journal should I submit to?" / "Find a journal" / "I was rejected, where else?"	/find-journal
"Sync submission" / "Retarget journal" / "Check SSOT drift"	/sync-submission
"Add a journal profile" / "저널 프로필 추가"	/add-journal
"Download PDFs" / "Get full texts" / "PDF 다운로드"	/fulltext-retrieval
"Visual abstract 만들어줘" / "Graphical abstract" / "GA 생성"	/make-figures
"De-identify my data" / "Remove PHI" / "비식별화" / "익명화" / "Anonymize patient data"	/deidentify
"Write an IRB protocol" / "Draft ethics submission" / "Research protocol"	/write-protocol
"Fill IRB protocol form" / "기관 양식 채워줘" / "심사면제 요청서 채움" / "동의면제 양식"	/fill-protocol
"Define my variables" / "Justify cutoff" / "Phenotype definition" / "변수 정의 근거" / "ad-hoc 정의 피하기"	/define-variables
"Write a case report" / "I have an interesting case"	/write-paper (case-report mode)
"Generate a cover letter" / "Write cover letter for submission"	/write-paper (Phase 8+, requires completed manuscript)
"Render manuscript to DOCX" / "Build final .docx" / "Cascade reformat references" / "Apply journal CSL" / "Re-render with Vancouver" / "회람용 docx" / "Zotero CWYW injection"	/manage-refs
"Render proposal to PDF" / "Anchor doc PDF" / "한글 학술 PDF" / "Briefing handout PDF" / "IRB cover PDF" / "non-bib markdown → PDF"	/render-pdf-doc
"Sync Zotero" / "Refresh refs.bib" / "Better BibTeX export" / "Zotero ↔ Obsidian 동기화" / "literature note 만들어줘"	/lit-sync
"PDF를 Obsidian 노트로" / "논문 요약 노트 만들어줘" / "이 폴더의 PDF 정리해줘" / "build a paper vault" / "second brain" / "extract concepts from papers"	/obsidian-paper-vault
"ICMJE COI 양식 일괄 생성" / "공저자 COI 폼 만들기" / "Disclosure form batch"	/fill-icmje-coi
"AI search optimization" / "Perplexity-friendly abstract" / "RAG visibility" / "GEO checklist" / "Elicit/Consensus 노출"	/academic-aio
"Remove AI patterns" / "AI 티 제거" / "humanize this section" / "GPT 흔적 지워줘"	/humanize

Multi-skill workflows (plan then execute sequentially)

The Nodes column lists the decision forks to render in interactive mode (Dialogue Protocol below); N1–N11 are defined in ${CLAUDE_SKILL_DIR}/references/dialogue_nodes.md.

Scenario	Skill chain	Nodes
New project, no prior work	intake-project -> search-lit -> design-study -> manage-project init	N1, N2 (if user wants manuscript output), N3
Data ready, need a paper	manage-project init -> analyze-stats -> make-figures -> write-paper	N6 (PHI gate), N3, N4 (journal timing), N2
Draft exists, prepare for submission	self-review -> check-reporting -> verify-refs -> humanize -> academic-aio (opt-in --aio) -> manage-refs (DOCX build + xref QC --strict) -> manage-project checklist	N4 (if not yet locked), N8 (only if self-review returns fatal)
Submission rendering & cascade reformat	manage-refs (Workflow A pandoc citeproc, or B Zotero CWYW) -> manage-refs scripts/check_xref.py --strict -> verify-refs -> sync-submission build	N10 (Workflow A vs B selection — see manage-refs SKILL.md decision tree)
Cascade rejection re-target	find-journal (exclude rejected) -> manage-refs (re-render with new CSL) -> write-paper Phase 8+ (new cover letter) -> sync-submission build --journal {new}	N4
Non-bibliography academic deliverable (proposal/briefing/anchor doc)	write-protocol -> fill-protocol (institutional form available) ⫶ render-pdf-doc (markdown-only, no form)	N11 (form available vs not)
Reference housekeeping cycle	lit-sync (Zotero ↔ refs.bib auto-export) -> manage-refs scripts/check_citation_keys.py -> verify-refs --strict (first-author cross-check)	—
ICMJE COI batch (multi-author submission)	fill-icmje-coi (per-author docx generation from synthetic seed) -> manual circulation	—
Medical-AI paper, AI-search visibility pass	self-review -> humanize -> academic-aio (title, abstract, summary box, README / CITATION.cff / HF card)	N4, N9 (section entry for re-edit scope)
Reviewer comments received	revise -> analyze-stats (if new analyses needed) -> make-figures (if new figures needed)	N1
Meta-analysis from scratch	search-lit -> fulltext-retrieval -> meta-analysis (handles its own pipeline internally)	N2 (MA type), N5 (synthesis scope)
Grant writing	search-lit -> grant-builder	N2 (option 5)
Conference presentation	present-paper (handles its own pipeline internally)	N1
New study, need IRB protocol	search-lit -> design-study -> calc-sample-size -> define-variables -> write-protocol	N3, N2 (option 4 — protocol)
Observational cohort study (retro/screening/registry)	intake-project -> design-study -> search-lit -> define-variables -> write-protocol -> analyze-stats -> write-paper	N1, N2, N3
Data with PHI, need full pipeline	deidentify -> clean-data -> analyze-stats -> make-figures -> write-paper	N6 (mandatory), N3, N4
Data ready, need cleaning first	clean-data -> analyze-stats -> make-figures -> write-paper	N6, N3, N4
Full submission chain	write-paper -> self-review -> check-reporting -> find-journal -> write-paper (Phase 8+ cover letter) -> manage-project checklist	N4, N8 (if recovery triggered), N9 (on re-entry)
Post-rejection resubmission	find-journal (exclude rejected journal) -> write-paper (Phase 8+ new cover letter)	N4
Case report pipeline	search-lit (similar cases) -> write-paper (case-report mode) -> self-review -> check-reporting (CARE) -> find-journal	N2 (option 2), N4

Panel mode (/self-review --panel) is opt-in, never automatic. The chains above use single-pass self-review. Add --panel only when the user asks for a high-stakes final pass, because it spawns several reviewer agents plus an editor (several times the tokens); never enable it by default, including in --e2e. Never combine --panel with --fix: a panel diagnoses and prioritizes, and the fix loop is a separate pass.

Ambiguous requests (ask before routing)

If the intent is genuinely unclear, ask ONE clarifying question — never two in one turn. If you can make a reasonable inference, act on it and confirm; inside a pipeline, render the matching node instead of asking. Examples:

• "Help with my paper" -> Ask: "Do you want to start writing, review an existing draft, or respond to reviewer comments?"
• "What should I do next?" -> Check for project_state.json or STATUS.md in the working directory first. If found, read it and suggest the next logical step. If not found, ask what they're working on.

---

Workflow Execution — Dialogue Protocol (interactive default)

Without --e2e, a multi-skill chain runs one decision node per fork. Never replace a node with a plan list and "Shall I proceed?", because that silently commits the paper type, study design, target journal or recovery branch. Also pause at write-paper's built-in gates (outline approval, discussion planning).

Load ${CLAUDE_SKILL_DIR}/references/dialogue_nodes.md the first time the pipeline enters a decision fork in the current session; it holds each node's options, rendering template and autonomous default.

Per-fork execution sequence:

1. Identify the node for the fork (Nodes column of the Multi-skill workflows table).
2. Render it with the reference's §"Rendering Template": under ~15 lines, unlocks / locks / recovery_cost per option, autonomous default announced.
3. Wait for a numeric choice or a control word. One node per turn — never stack two.
4. Echo the lock in one line before invoking the skill ("Locking: CARE reporting guideline; abstract = structured 250w.").
5. Invoke the downstream skill for the chosen option, then return to step 1 for the next fork.
6. Adapt on skill output. If a result invalidates a prior lock (e.g., /self-review surfaces a Step 7.4a trigger), route to the recovery node (N8) instead of continuing the chain.

Control words. back re-enters the previous node. pause halts the pipeline and returns control to the user. skip is allowed only for a node whose locks scope is empty (rare) — otherwise explain why skipping is not available.

---

Full Pipeline Mode

When the user asks to "run the full pipeline" or similar, run the Standard Pipeline below.

--e2e Flag

Pre-flight Validation (run once at --e2e entry)

Before invoking any downstream skill in --e2e mode, run the following 4 checks. A failure on any one halts the pipeline. Write all four outcomes verbatim to manuscript/<id>/REPORT.md (see §"REPORT.md Generation") under Frozen / Version status + Source artifacts checked.

1. STATUS / project_state: read STATUS.md or project_state.json in the working directory and confirm the current phase. If neither exists, halt with STATUS_MISSING unless the user passes --no-status.
2. Frozen artifact: scan manuscript/<id>/v_*_package/. If the latest v_N carries a _FROZEN marker file or INDEX.md::frozen=true, this run is restricted to a v_(N+1)_package/ branch. Any attempt to write directly into v_N halts with FROZEN_VIOLATION.
3. Required inputs: confirm input artifacts for the requested phase exist. Examples: Phase 4 figure requires analysis/_analysis_outputs.md; Phase 7 self-review requires manuscript/manuscript.md. Missing → halt with REQUIRED_INPUT_MISSING: <path>.
4. Dependency miss: if the user requested phase k but a prior phase is incomplete, halt with DEPENDENCY_MISS: [Phase i, Phase j] by default. Only when the user explicitly passes --auto-extend may the orchestrator prepend the missing phases and continue.

--e2e Pipeline Behavior

When --e2e is passed (or the user says "end-to-end", "Arm A", or "fully autonomous"):

1. Set --e2e mode ON.
2. Pass --autonomous to /write-paper when invoking it.
3. Pass --json to /self-review and /check-reporting when invoking them.
4. Skip all orchestrator-level confirmations ("Shall I proceed?") and do NOT render any Dialogue Protocol nodes.
5. For each node the pipeline would have rendered interactively, apply the node's default (from dialogue_nodes.md) and log the choice to qc/_pipeline_log.md as: [orchestrate] N{id}: defaulted to option {n} ({label}) — {autonomous_rationale}.
6. DO still respect data-safety gates (PHI Safety Gate / node N6): if PHI status is unknown, HALT the autonomous run with a single prompt. After pre-flight passes, PHI is the only node that can interrupt autonomous mode.
7. Audit Recovery (node N8): auto-invoke the routed recovery skill. If the route itself fails validation twice, HALT with RECOVERY_HALT_HUMAN_DECISION in the log.
8. AIO (academic-aio) is OFF by default in --e2e, because AI-search visibility is a pre-submission concern and running it on every autonomous iteration would waste tokens and invite silent rewrites that violate the skill's "never edit silently" contract. Enable it only when the user explicitly adds --aio (or the pipeline is preparing a preprint / GitHub README / HF card alongside submission). When enabled, schedule it after /humanize so the checklist anchors on QC-confirmed and human-readable text, and surface the PASS/PARTIAL/FAIL report to the user — never auto-apply its edits.
9. After each skill completes, run post-skill validation (see below); once it passes, proceed to the next skill without asking.

Standard Pipeline: Data → Manuscript

1. /analyze-stats → analysis/tables/*.csv, analysis/figures/*, analysis/_analysis_outputs.md, analysis/analyze.py
2. /make-figures --study-type {type} → reads analysis/_analysis_outputs.md → analysis/figures/*.pdf, analysis/figures/*.png, analysis/figures/_figure_manifest.md
3. /write-paper --autonomous (if --e2e) → reads analysis/ → manuscript/manuscript.md (DOCX rendering delegated to step 7)
   • Phase 7.4 internally calls /self-review --json --fix → qc/self_review.md
4. /check-reporting → reads manuscript/manuscript.md → qc/reporting_checklist.md + qc/reporting_checklist.json (Part D JSON; called within write-paper Phase 7, but orchestrator verifies output)
5. /verify-refs → reads manuscript/manuscript.md → qc/reference_audit.json (sole output; row-level status in records[])
6. /self-review --json --fix → reads manuscript/manuscript.md → qc/self_review.md + auto-fix (called within write-paper Phase 7.4, but orchestrator verifies final output)
7. /manage-refs (Workflow A pandoc citeproc, or B Zotero CWYW) → reads manuscript/manuscript.md + manuscript/_src/refs.bib → manuscript/manuscript_final.docx + qc/xref_audit.json. Submission gate: check_xref.py --strict must pass (no MISSING_DOCX / MISSING_BODY / MISMATCH).

Post-Skill Validation

After each skill completes, verify that expected output files exist, then pass the discovered file paths to the next skill as context. If validation fails, report the error and do NOT proceed to the next skill.

Skill	Expected Outputs	Validation
/analyze-stats	At least one file in analysis/tables/*.csv OR analysis/_analysis_outputs.md	Check file existence and non-empty
/make-figures	analysis/figures/_figure_manifest.md with at least 1 entry	Parse manifest, verify listed files exist
/write-paper	manuscript/manuscript.md (required)	Check file existence and non-empty. Do NOT require the DOCX here — manuscript_final.docx is rendered later by /manage-refs (step 7), so requiring it would halt an --e2e run before the DOCX exists
/check-reporting	qc/reporting_checklist.md and qc/reporting_checklist.json (Part D JSON)	Parse the JSON; halt naming the file if it is absent, unparseable, or lacks an integer missing. If missing > 0 or any action_items[].status == "MISSING", halt with REPORTING_ITEMS_MISSING (list each item) and do NOT proceed to step 7 — route the gaps to /write-paper Phase 7. An inline report that wrote no file does not pass
/verify-refs	qc/reference_audit.json (sole output)	Parse JSON; halt if submission_safe == false (i.e., FABRICATED / MISMATCH count > 0 OR duplicate_findings[] nonempty)
/self-review	qc/self_review.md with the Phase 3c JSON block (--json)	Parse the JSON block; halt naming the file if it is absent, unparseable, or lacks an integer fatal_count. If fatal_count > 0 or any issues[].severity == "fatal", do NOT proceed to step 7 — route to N8 (Audit Recovery). A REVISE verdict with no fatal issue is logged (score, verdict, counts) in qc/_pipeline_log.md and REPORT.md, not halted. Accept the optional consensus array and R1/R2/R3 attributions that --panel adds to issues (additive, backwards-compatible)
/manage-refs	manuscript/manuscript_final.docx, qc/xref_audit.json	DOCX exists and non-empty; xref_audit.json has submission_safe: true (no P0 blocker rows)
/lit-sync	manuscript/_src/refs.bib (mtime updated), references/zotero_collection.json	refs.bib mtime newer than collection snapshot; refs_bib_refreshed: true in collection JSON

On validation failure:

• Log the failure: which skill, which output was missing, any error messages.
• In --e2e mode: report the error in qc/_pipeline_log.md and STOP. Do not proceed to the next skill. Output: "Pipeline halted at {skill}: {missing output}. Check the skill's output and re-run."
• In interactive mode: report the error and ask the user how to proceed.

REPORT.md Generation

At the termination of every --e2e invocation — whether the pipeline completed, halted at pre-flight, or halted on post-skill validation — the Worker MUST write manuscript/<id>/REPORT.md using the template at ${CLAUDE_SKILL_DIR}/references/report_template.md.

• Copy every section of the template verbatim. Never delete a section. Empty fields are filled with (none) or (unknown) — never omitted, never collapsed.
• The §"Pipeline log" entry is a 5-line summary of qc/_pipeline_log.md (Dialogue node defaults applied, skill invocations, halt reason if any) — not a paste of the full log.
• The §"Tier-3 Blocked Items" hook-vs-prompt-guard split is mandatory — see §"Tier-3 Worker Guard" below.
• The §"Next safe command" line is the literal command the user can copy to resume the next phase. Do not editorialize.
• REPORT.md is the single artifact the user reviews; every other QC output is linked from it.

Tier-3 Worker Guard

The following actions are permanently forbidden inside --e2e autonomous flow. On detection, the Worker halts the pipeline and records the attempt under REPORT.md §"Tier-3 Blocked Items" as tier3_pending: <command>. Hook-confirmed blocks and prompt-only blocks are listed separately so a future hook regression cannot silently re-open a prompt-only block.

Hook-confirmed (where a Tier-3 confirm hook is installed):

• gws gmail +send / +reply
• YouTube upload

Prompt / skill guard only (no hook coverage — Worker prompt enforces):

• git push, gh pr create
• MCP Gmail send, MCP Calendar send
• MCP GitHub create-pr
• /sync-submission build external publication paths
• Phase 8 submission DOCX auto-build / journal submission
• Senior mentor automatic email reply

git commit is allowed; a subsequent git push attempt halts. Circulation emails are saved as a Gmail Draft only — never sent.

Data Flow Contract

Read ${CLAUDE_SKILL_DIR}/references/data_flow_contract.md when chaining a skill and you need what it reads or writes (the Standard Pipeline above already names its own files).

Post-E2E: Journal Selection & Submission Prep

After the E2E pipeline completes (or when the user requests journal targeting), this workflow is available. It is NOT part of --e2e: it runs only on explicit user invocation, because it needs the user's journal selection.

1. /find-journal → top 5 recommendations based on manuscript/manuscript.md abstract
2. /verify-refs → block fabricated or mismatched references before packaging
3. User selects a journal → create submission/{journal_short}/ directory
4. /sync-submission build --journal {journal_short} → create or refresh the derived manuscript package from the canonical manuscript
5. Generate inside submission/{journal_short}/:
   • cover_letter.md: via /write-paper Phase 8+
   • checklist.md: journal-specific submission checklist
   • manuscript_final.docx: re-rendered by /manage-refs for the target journal (if format differs)
6. /self-review with the selected journal as the target → journal scope-aware final pass (not /peer-review, which reviews other authors' work only)

---

PHI Safety Gate

Before routing to any data-handling skill (clean-data, analyze-stats, make-figures), check if the data might contain PHI:

1. For each CSV/Excel file the task will read that is not itself a *_deidentified.* output (a de-identified copy of one file says nothing about the others), ask: "Does the data contain patient identifiers (PHI)? (names, national ID / RRN, date of birth, contact details, etc.)"

   • If yes → Route to /deidentify first, then continue to the originally requested skill using the *_deidentified.* output file
   • If no, or the user confirms the data is already de-identified → Proceed directly

2. De-identification is an INTERACTIVE process requiring the researcher's active participation. Warn: "De-identification requires the researcher's direct review. You must run the script in the terminal and verify each item."

---

Context Detection

Before routing, check for context clues in the working directory:

File found	Implies
project_state.json	Active managed project -- read it to determine current phase
STATUS.md	Project with status tracking -- read current stage and blockers
PROJECT.md	Project identity exists -- use for context
CLAIMS.md	Claims-to-results map exists -- writing is underway
REVIEW_LOG.md	Revision cycle -- likely needs /revise
*.qmd or *.tex files	Manuscript drafting in progress
*.bib files	References exist -- may need verification
PRISMA_*.md or QUADAS*.md	Meta-analysis or systematic review
Decision letter / reviewer PDF	Route to /revise
CSV/Excel data files without analysis scripts	Raw data may need cleaning -- suggest /clean-data first
*_deidentified.* or audit_log.csv	Some data was de-identified -- skip the PHI Safety Gate only for the *_deidentified.* outputs themselves; run it on every other data file the task will read
protocol_draft.md	Protocol drafting in progress -- may need /write-protocol
sample_size_*.csv or sample_size_*.R	Sample size calculation done -- check if protocol or manuscript next

---

Guardrails

• Never invent a skill. Only route to skills listed in the Available Skills table.
• Respect existing state. If a project scaffold exists, do not re-initialize it.