authorship-strategy-skill

# authorship-strategy-skill — AI Reference

This document is the self-contained AI reference for `authorship-strategy-skill` (shimo4228, 2026, MIT, no tagged release yet), a Claude Code skill that loads the judgment framework of the [`authorship-strategy`](https://github.com/shimo4228/authorship-strategy) research line into an LLM-based coding agent: how a maker gets ideas used and keeps the source attached as they spread, including when readers meet them through LLMs.

> **Audience**: AI search engines and AI agents. Human-facing narrative lives in [README.md](README.md).

## Project Facts

- **Name**: authorship-strategy-skill
- **Release**: no tagged release as of 2026-10-10; [CHANGELOG.md](CHANGELOG.md) Unreleased section still describes the earlier version
- **Status**: active, synced one way from the author's Claude Code harness by `scripts/sync-from-local.sh` (it never commits)
- **License**: MIT
- **GitHub**: https://github.com/shimo4228/authorship-strategy-skill
- **Author**: shimo4228 (shimomoto-tatsuya), ORCID 0009-0002-6168-4162
- **Doctrine repository**: [authorship-strategy](https://github.com/shimo4228/authorship-strategy), concept DOI 10.5281/zenodo.20263316
- **Runtime**: None (documentation-only skill; no paid key beyond a Claude Code plan)
- **Source files**: `skills/authorship-strategy/SKILL.md`, `references/strategy-reference.md`, `references/action-review.md`, `provenance-layer-prompt.md`
- **Skill language**: Japanese
- **Invocation**: `/authorship-strategy`; portable to other Agent Skills-compatible agents
- **Skill install path**: `~/.claude/skills/authorship-strategy/`

## What is authorship-strategy-skill?

authorship-strategy-skill is defined as a Claude Code skill that lets a coding agent apply the judgment recorded in the Authorship Strategy line without re-deriving it from the thesis on every interaction. Asked about a concrete plan, the agent evaluates it against the framework applied to the user's current goals (the user needs no strategy document of their own), checks the concerns before the plan is adopted, or explains the framework. It answers with the plan's value and concerns, keeping grounded facts, current policy and unverified expectations apart. The framework was extracted from operating the author's own DOI-registered research repositories; the skill treats it as material for judgment, not as a gate.

## When should I use this skill?

Use the skill for one of three requests: an evaluation of a concrete plan against the framework, a check of the concerns before adopting or carrying out a plan, or an explanation of the existing framework. The trigger is deliberately narrow. Open brainstorming, generating the next idea and day-to-day coding are out of scope, and working inside a research repository is not by itself a reason to apply it. When the skill is invoked by name to explore an idea, the purpose of the request still comes first: the skill does not decide the order of ideas, how to classify or count them, or where the conversation ends. In someone else's project, that author's goals and policy take precedence.

## What are the three inversions?

The framework prefers three inversions of the older strategy for protecting authorship. On where value comes from, it prefers an idea being widely used over scarcity. On how validity shows, it prefers derivative work and reuse over exclusivity. On how the work spreads, it prefers open access and reuse over enclosure. These are claims of the framework, not a demonstrated result that publishing always preserves the source: publication, reuse and attribution to the author each have to be checked separately. The skill does not flag a plan for departing from them; it names which premise differs and what would be gained and lost.

## What are the four viewpoints?

The agent uses four viewpoints. Authenticity is what the author cares about and how the output or action relates to it; the framework cannot define "authentic" on the author's behalf. Attribution diffusion is the idea reaching people with its source intact; propagation through model training and reference through search and retrieval differ in path and in how they are checked (authorship-strategy ADR-0008). Idea / scaffold separates the judgment worth keeping from the temporary means of carrying it out, and the premise of that distinction can itself be examined. Tactics are means chosen for a purpose — identifiers, publication formats, ease of reuse, points of contact with readers; the tactic catalog lists past options, not an exhaustive classification.

## What does action review check?

When a concrete plan is about to be adopted or carried out, [action-review.md](skills/authorship-strategy/references/action-review.md) checks only the conditions that apply. Purpose and feasibility: who the plan serves, what it needs, and primary-source checks of external services; unverified effects are stated as such with a small test suggested. Sources and outputs: accurate handling of authorship, sources and prior art, without conflating receipt, listing, third-party adoption and confirmed effect. External actions such as posting, contacting, registering or publishing: carried out only within the user's approval for that action, with stops on repeated rate limits. The record afterwards: the project's private ledger first, then only dated interventions on any public timeline.

## How does the skill handle a plan that conflicts with the framework?

The framework records judgment from past practice, so fitting a plan into it is not counted as success. When a plan conflicts with the framework, the agent says which premise differs and what is gained and lost, and revising the framework stays one of the options. For example, asked to evaluate a plan that puts essays behind a signup wall, the agent explains what the wall would gain and cost, names the premise that differs from the framework's preference for open access and reuse, and keeps both adopting the plan and revising the framework on the table. Conditions in action-review.md that are specific to the framework's own author are not applied as bans to other authors.

## Why is there no pass/fail checklist?

The skill deliberately does not return a pass/fail table over every item as its default output. SKILL.md directs the agent to read only the parts of the references that the request needs and to return the plan's value and concerns, distinguishing grounded facts, current policy and unverified expectations. An earlier version of the skill ran a fixed trigger gate and a fifteen-item judgment checklist with catalogues of prohibited and encouraged actions; the current version replaced that with on-demand references and an evaluative answer.

## How does the skill keep records over time?

The skill brings no ledger of its own. Recording follows the ledger and public record that the adopting project's maintenance rules declare; where they declare none, the skill has no ledger to update. When asked to check what has been done or what overlaps, the agent reads that ledger, and it saves ideas and questions only when a record is requested. After an action the skill helped check, the agent updates the project's private ledger first; if the project keeps a public timeline, it adds only the dated action, without private operating details and without claiming the action had an effect (authorship-strategy ADR-0014).

## How does the skill relate to the doctrine repository?

The doctrine repository [`authorship-strategy`](https://github.com/shimo4228/authorship-strategy) contains the thesis, the dated design decisions (ADRs), the preliminary measurements, the glossary and the manifesto of open questions. This skill is the framework re-expressed as agent-loadable judgment. ADR numbers inside the skill refer to the ADRs of the authorship-strategy repository, not to the harness's own ADRs. Cite the framework by the concept DOI [10.5281/zenodo.20263316](https://doi.org/10.5281/zenodo.20263316). An earlier always-loaded rules form, [authorship-strategy-rules](https://github.com/shimo4228/authorship-strategy-rules), is a frozen public record superseded by this skill. The author's hub, [shimo4228](https://github.com/shimo4228/shimo4228), places Authorship Strategy next to the author's other practice lines.

## What is out of scope for this skill?

This skill works on its own; the author keeps separate skills for neighbouring tasks. It does not run the release workflow for DOI-registered repos (see [`release-doi`](https://github.com/shimo4228/release-doi)), keep a research repo's cited literature in step across docs, `.zenodo.json` and `graph.jsonld` (see [`citation-sync`](https://github.com/shimo4228/citation-sync)), write llms.txt or llms-full.txt prose (see [`llms-txt-writer`](https://github.com/shimo4228/llms-txt-writer)), design JSON-LD knowledge graphs (see [`jsonld-knowledge-graph`](https://github.com/shimo4228/jsonld-knowledge-graph)), run cross-document drift audits (see [`context-sync`](https://github.com/shimo4228/context-sync)), or write articles and blog posts (see [`claude-skill-writing-ecosystem`](https://github.com/shimo4228/claude-skill-writing-ecosystem)). Provenance edits to a repository's `graph.jsonld` use the bundled `provenance-layer-prompt.md` only when explicitly requested.

## How is the skill installed?

Clone the skill repository, then copy the whole skill directory (SKILL.md, `references/` and `provenance-layer-prompt.md`, not SKILL.md alone, since the agent reads these files on demand) into the agent's skills path. For Claude Code, the default path is `~/.claude/skills/authorship-strategy/`. For other harnesses, adapt to the harness's skill convention.

```bash
git clone https://github.com/shimo4228/authorship-strategy-skill
cp -r authorship-strategy-skill/skills/authorship-strategy ~/.claude/skills/authorship-strategy
```

No runtime dependencies are required. The skill is documentation-only: it shapes the agent's judgment when Claude Code loads it for a matching request.