release-doi
shimo4228/release-doi
Claude Code skill: release workflow for DOI-registered research repositories following the authorship-strategy identifier-federation triplet (ADRs 0001-0003). Five-phase verify-and-deposit runbook.
Instalação
npx convez-ai shimo4228/release-doi1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
release-doi
A Claude Code skill that runs the release workflow for DOI-registered research repositories following the identifier-federation triplet (ADRs 0001-0003) of the authorship-strategy research line. Sequences pre-release verification, tag-push, archive deposit, DOI propagation, and cross-platform federation update steps as a single five-phase runbook that prevents the most common release-time drift incidents (off-by-one canonical-reference mistakes, missed sibling cross-reference updates, accidental version-DOI canonicalization).
The skill is harness-aware where it has to be (it knows about Git, GitHub, Zenodo's auto-deposit-on-tag webhook, and CITATION.cff / .zenodo.json formats) but harness-neutral where it can be: the verify and deposit steps are written so an adopter using a different archive service (Software Heritage, OSF, Figshare) or a different release tooling chain can substitute the equivalent operations without re-deriving the runbook.
When to use
Apply the skill when all of the following hold:
- The repository is registered (or about to be registered) with a versioned DOI archive (Zenodo is the reference example)
- The release follows a tagged-release-triggers-deposit model (typical for Zenodo's GitHub integration)
- The repository participates in a federation of sibling DOI-registered artifacts (
.zenodo.jsonrelatedIdentifiersdeclarations, cross-platform dataset mirrors) - The release author intends to maintain the concept DOI as canonical reference discipline (ADR-0001 in the upstream research line)
Skip the skill for:
- Repositories without DOI registration (the workflow's central artifact, the DOI, does not exist)
- One-shot artifacts published without versioning intent (the version/concept DOI distinction does not apply)
- Repositories where the release author has explicitly opted out of the identifier-federation triplet (the skill's assumptions do not hold)
Install
Claude Code
git clone https://github.com/shimo4228/release-doi
cp -r release-doi/skills/release-doi ~/.claude/skills/release-doi
No runtime dependencies for the skill itself; verification commands inside use git, optional python3 -m json.tool for JSON-LD validation, and optional cffconvert for CITATION.cff syntax checking. The skill invokes the harness's release tooling (git tag, gh release create or equivalent) rather than wrapping them.
Other harnesses
Adapt the install path to your harness's skill convention. The skill's runbook is written in a form that does not depend on Claude Code specifically; the trigger and invocation mechanism is the only harness-specific surface.
How it works
The skill walks through a pre-flight gate, five sequential phases, and one post-release phase. Each phase has an explicit completion condition; the skill refuses to advance until the condition is met.
- Pre-flight: Zenodo opt-in check — verifies the Zenodo GitHub webhook is registered on the target repository (
gh api repos/<owner>/<repo>/hooks). Zenodo's GitHub integration is repo-by-repo opt-in; releases cut before the toggle is enabled are not retroactively archived. This gate catches the gap on first release of a new DOI repository — a blind spot when sibling repos are already opted in. - Phase 1: Baseline — capture the pre-release ground truth (commit count, file count, test count where applicable). This is the reference state against which Phase 4 verifies.
- Phase 2: CODEMAPS regeneration — re-generate file-level architecture maps so they reflect the to-be-released state.
- Phase 3: Cross-document consistency — sync CHANGELOG / CITATION.cff /
.zenodo.json/pyproject.toml(where applicable) / multilingual README / llms.txt / glossary / ADR bidirectional links. Drift detection delegated to context-sync where available. - Phase 4: Verify — CITATION.cff syntax, version triple consistency, lint and test pass, secret scan, unintended-file inclusion check.
- Phase 5: Release execution — stage specific files only (no wildcards), HEREDOC commit message, tag creation, push main and tag, create the Git host release object that triggers the archive webhook.
- Post-release — wait for the archive's DOI minting (typically five to ten minutes for Zenodo), verify the concept DOI is what gets propagated (not the initial version DOI; the off-by-one drift incident this guards against is documented in ADR-0001 of the upstream research line), then update the CITATION.cff and
.zenodo.jsonwith the concept DOI and create a follow-upchore(release):commit.
Key concept: concept DOI vs version DOI
The skill's most load-bearing single discipline is the distinction between the concept DOI (parent record, resolves to the latest version, the canonical reference shape) and the version DOI (specific to one release, used only for reproducibility citations). At initial deposit the two are typically issued as adjacent integers, and a user copying the citation snippet from the just-published version's UI page usually copies the version DOI by mistake. The skill's post-release phase explicitly asks the operator to verify which DOI is being propagated and to back out and correct if the version DOI was used.
This is not a hypothetical concern: the upstream research line's ADR-0001 was extracted from an actual sixteen-file drift incident across the author's own sibling repositories. The skill exists in part to prevent that incident from recurring.
What this skill does NOT do
| Concern | Use this instead |
|---|---|
| Decide whether to apply the identifier-federation triplet at all | authorship-strategy-skill — applies the framework's trigger conditions |
| llms.txt / llms-full.txt prose | llms-txt-writer |
| JSON-LD knowledge graph design | jsonld-knowledge-graph |
| Cross-document drift detection (Phase 3 delegates to this) | context-sync |
| File-level architecture map regeneration (Phase 2 delegates to this) | A update-codemaps skill or comparable, where available |
Related research and skills
- Doctrine repository: authorship-strategy — the normative framework, five tactical ADRs (especially the identifier-federation triplet 0001-0003), and empirical baseline this skill is the operational instantiation of
- Peer components (other component skills of the same framework):
- authorship-strategy-skill — the framework's judgment-checklist form; this release-time skill assumes the framework has already determined the artifact is in-scope
- llms-txt-writer — operationalizes Layer 4 tactic 7's Answer.AI
llms.txtconvention; this release skill's Phase 3 invokes it when llms.txt regeneration is needed - jsonld-knowledge-graph — operationalizes Layer 4 tactic 7's JSON-LD knowledge graph; this release skill's Phase 4 uses its verification commands
- Sibling research lines (at the research-program level): Agent Knowledge Cycle (AKC), Contemplative Agent, Agent Attribution Practice (AAP)
Terminology note. This ecosystem reserves sibling for research-line-level peers; at the component-skill level the term peer component is used instead.
License
MIT. See LICENSE.