Date: 24 September 2026. Environment: Linux x86_64, Python 3.10.12, Codex CLI 0.154.0.
Version: 1.0.0-public.1. Archive: biomedical-literature-evidence-1.0.0-public.1.zip.
SHA-256: a5d633dd66272876b16678557f2941fb06cdbce1e8dd364326c4355d68a54241.
Contents: 33 files, 211893 bytes unpacked.
Executed checks
- Frozen archive entries match every source path, byte count and SHA-256. Safe extraction and relative Markdown links passed; Python and JSON parsed successfully. Skill frontmatter validation passed.
- From a fresh temporary extraction: 19 unittest methods passed (exit 0).
- Native Codex discovered the project-local public entrypoint from the final ZIP, read the correct version and executed the PubMed retrieval help command with exit 0.
- Both synthetic reference imports retained three records and left all identities unverified. A synthetic insufficient ledger returned completed=true and support_ready=false.
- Live provider: the title-query warning was reported as partial; PMID lookup, metadata fetch and PMC BioC acquisition succeeded. The log contains UTC times and raw-response hashes. Raw article text is not redistributed.
Reproduce locally after obtaining the ZIP
From the extracted skill directory:
PYTHONDONTWRITEBYTECODE=1 python3 tests/ledger_dev.py python3 scripts/pubmed_retrieve.py --help
To repeat the known-record lookup, run python3 scripts/pubmed_retrieve.py --raw-dir NEW_RAW_DIRECTORY search --query "35939577[PMID]" --max-records 5 --batch-size 5, then use fetch 35939577 and fulltext PMC9716639 with separate --output paths. Live results may change. Import the linked synthetic library with scripts/import_records.py INPUT --format ris or --format csl-json. Validate the linked synthetic ledger with scripts/review_ledger.py LEDGER.
Review and limitations
Owner-authorized publication; AI-assisted preparation and self-review. No independent human review or blinded scientific quality score is claimed. Native-host checks cover discovery and the stated command only. Claude Code, Antigravity, macOS and Windows invocation were not tested. Biomedical retrieval requires POSIX file locking; native Windows is unsupported.
The scientific example is one known-report reading, not a systematic review. Supplements, participant data and a comprehensive publication-notice search were not inspected. Ledger checks recorded accounting and attestations rather than independent semantic truth. Historical 103-assertion and 48-provider-test totals were not rerun.
An unrelated preinstalled connector reported an authentication startup error during the Codex check. It was not used by the tested local commands, which passed; no connector credentials or settings were changed.