Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.4.0] - 2026-03-23

Step 3 record-keeping: immediate write-to-file — metadata from get_article_metadata must now be written to _working_ledger.yaml immediately after each call, in the same turn, not batched. This prevents context drift from corrupting DOIs and author names.
Step 4 verification: verify ALL references, not spot-check — previously said "spot-check at least 5"; now requires re-fetching metadata for every PubMed reference and comparing DOI, title, and first author character-by-character. PubMed is authoritative; any mismatch overwrites the ledger.
Step 5: re-read ledger from disk — before writing the reference list, the model must re-read _working_ledger.yaml from file, not from context memory.
YAML schema: expanded author fields — authors replaced with first_author (surname + initials) and authors_full (complete author list), both copied verbatim from PubMed.
Added real failure case as concrete warning in record-keeping section (DOI differed by 3 characters, wrong first author — caught only by re-verification against PubMed).

bioRxiv/medRxiv integration (Step 3b) — searches for preprints on medRxiv/bioRxiv when the bioRxiv MCP is available; results go in a separate "Emerging Evidence" section clearly labeled as non-peer-reviewed
ClinicalTrials.gov integration (Step 3b) — searches for ongoing/completed trials when the Clinical Trials MCP is available; results appear in the Evidence Gaps section with NCT numbers, phase, and expected completion dates
Incremental ledger writing — reference ledger is now written to _working_ledger.yaml after each batch of searches to protect against context loss during long searches
Retraction checking — Step 4 verification now checks for retracted papers in PubMed and excludes them
Scholar Gateway DOI verification — Scholar Gateway-only references now get cross-checked against PubMed in Step 4; removed the claim that Scholar Gateway DOIs are "reliable"
Scope guardrails — Step 1 now includes a scope check; topics that are too broad or too narrow get flagged before searching begins
Conflicting Recommendations section — new template section (Section 4) for cases where guidelines disagree or recent evidence contradicts guidelines
BibTeX key collision handling — uses AuthorYeara, AuthorYearb convention when multiple papers share the same key
New eval: CAR-T therapy (tests bioRxiv and ClinicalTrials.gov integration)
New eval: overly broad topic (tests scope guardrail)

find_related_articles guidance: now recommends using on 2-3 landmark papers only, not every paper
YAML schema: added preprints and ongoing_trials sections (optional)
Evidence summary template: renumbered sections (4=Conflicting Recommendations, 5=Emerging Evidence, 6=Evidence Gaps, 7=Transparency Disclaimer, 8=References)
Version string no longer needs dynamic propagation — hardcoded in YAML schema, removed from transparency disclaimer
Model requirements section renamed to "How to Approach This Workflow" — model recommendation moved to README

Scholar Gateway-only references previously had no DOI verification (Step 4 was PubMed-centric)
Version string fragility — model no longer needs to read frontmatter version and propagate to two locations

Removed protocol-review mode — skill now has a single output mode (standalone evidence summary)
Protocol-reviewer skill now auto-triggers literature-search when no YAML is provided, removing the need for a separate mode
Removed protocol_section field from YAML ledger schema (protocol-reviewer maps references to sections itself)
Removed protocol-review mode template from evidence_summary_template.md

Protocol-review mode now detected from natural language (mentions of "protocol", "protocol review") instead of --protocol flag
Removed duplicated document structure from SKILL.md — now references evidence_summary_template.md as single source of truth
Added Quick Start section to SKILL.md
Added error handling guidance for missing tools (pandoc, PubMed MCP, Scholar Gateway)
Added .gitignore for generated output files
Added .claude/settings.local.json with pandoc permissions

6-step clinical literature search workflow (SKILL.md)
PubMed MCP and Scholar Gateway dual search strategy
Evidence summary document generation (.md → .docx via pandoc) with two modes:
- Standalone mode: full document with title page for general use
- Protocol-review mode (--protocol): reformatted for protocol-reviewer skill
YAML reference ledger output (_References.yaml) as contract for protocol-reviewer
BibTeX (.bib) and PMID list (.txt) export for Zotero import
PubMed search strategy guide (references/pubmed_strategy.md)
Evidence summary template (references/evidence_summary_template.md)
Reference integrity verification step (DOI spot-check, title-PMID consistency)
Draft callout and transparency disclaimer in every evidence summary
Evaluation scenarios for CMV prophylaxis, perioperative anticoagulation, and ABOi transplant

Split from protocol-reviewer v1.4.0
The literature search steps (formerly Steps 2-4, 7 of protocol-reviewer) are now a standalone skill
Evidence summary document is a new output not present in the original protocol-reviewer