README quality blueprint
RC1 planned public routes cover en and zh-Hant-TW; the 41-locale source catalog is private. This editorial blueprint remains canonical in English.
A Better Workflows README is a landing page, not a compressed reference manual. Its job is to help a reader answer five questions in order:
- What is this, and is it for me?
- What problem does it solve?
- Why should I trust its claims?
- What is the shortest path to a first success?
- Where should I go next?
This blueprint defines the narrative, visual, localization, and validation contract for every repository README. The machine-readable source is readme-quality-v1.json.
Start from the reader's decision
GitHub surfaces a README before most repository content. The first screen must therefore establish the product promise, intended audience, and a bounded next action. It must not begin with internal architecture, a complete command reference, or release-recovery detail.
Write for these reader jobs:
- New visitor: decide quickly whether Better Workflows solves a relevant problem.
- New user: install the plugin and reach one successful automatic route.
- Evaluator: understand the authority boundary and fail-closed behavior.
- Returning operator: jump to a workflow, security, architecture, or CLI answer.
- Contributor or translator: find the canonical contract, development commands, support, and governance.
Use a cause-and-effect narrative
The five landing READMEs use the same eight-part semantic sequence. Headings may be idiomatic in each language, but the reader journey does not change.
| Section | Reader question and narrative role |
|---|---|
| Promise and audience | What is Better Workflows, why does it exist, and who is it for? |
| Problem to outcome | What goes wrong when intent, authority, evidence, and provider outcome are conflated? |
| Proof and boundaries | Which guarantees make the proposed outcome credible? |
| First success | What is the shortest complete install-to-result path? |
| Choose the next path | Which workflow or document matches the reader's goal? |
| Lifecycle | How does a goal become a reconciled completion—or stop safely? |
| Trust and limits | What can the system never infer, authorize, or claim? |
| Learn, help, contribute | Where are the deep docs, support, governance, development, and license? |
This order provides a practical arc:
- Context: prompt-driven work can express intent without proving authority or state.
- Tension: side effects turn that gap into delivery risk.
- Resolution: Better Workflows binds goal, scope, evidence, review, action, and provider reconciliation.
- Proof: explicit guarantees and boundaries show how the resolution works.
- Action: the reader reaches a first success before encountering deep implementation detail.
- Continuation: role- and outcome-based routes move the reader to the right tutorial, how-to guide, explanation, or reference.
Separate landing content from deep documentation
Use the README for decision-relevant information. Route depth by purpose:
- Getting started is the first-use tutorial.
- Workflows is the outcome-selection how-to guide.
- Architecture explains the control plane and trade-offs.
- Security explains authority, privacy, attestations, and fail-closed behavior.
- CLI reference is the command reference.
- Localized
docs/details/*.mdpages preserve comprehensive translated detail.
Do not duplicate cache recovery, lock ownership, full provider transport semantics, exhaustive commands, or implementation change history in the landing page. A concise safety claim stays local; its auditable depth belongs in the canonical guide.
This separation follows the Diátaxis distinction between tutorials, how-to guides, explanation, and reference. A single page cannot optimize for all four reader needs at once.
Make every visual earn its place
Use a visual only when relationships, hierarchy, or state transitions are materially easier to understand than prose.
The landing pages permit two visuals:
- Authority-boundary architecture: answers which layers shape intent, current facts, tool authority, bounded retries, and read-only state.
- Goal-to-completion lifecycle: answers where evidence is checked, where side effects are authorized, and where unknown state stops progress.
Every visual must include:
- concise, meaningful alt text;
- an adjacent text equivalent that preserves the conclusion when the visual is hidden or Mermaid does not render;
- real text for essential labels whenever possible;
- a stable reader question that justifies keeping the visual current.
Do not add decorative screenshots, text-heavy images, or a diagram that merely duplicates a short list. Keep selector tables to two concise columns so they remain usable on narrow screens.
Preserve meaning across languages
English is the semantic reference, not a line-count target. Traditional Chinese, Simplified Chinese, Japanese, and Korean should sound natural to a native reader while preserving the same contract.
The following items must remain equivalent:
- the eight semantic sections and their order;
- first-success commands and product identifiers;
- the five authority, evidence, unknown-state, prompt, and privacy claims;
- workflow, security, architecture, CLI, support, governance, development, and license destinations;
- visual purpose, lifecycle stages, and text fallbacks;
- version source and badge policy.
Headings, sentence boundaries, punctuation, examples, and calls to action may be idiomatic. Never translate commands, selectors, evidence identifiers, or security semantics.
Write for scanning and translation
- Lead with the reader's outcome and put important terms at the beginning of headings and paragraphs.
- Use active voice and name the actor responsible for an action.
- Address the reader directly for procedures.
- Keep paragraphs short and give each one a single job.
- Use numbered lists for sequence and bullets for non-sequential choices.
- Use descriptive links instead of generic “click here” labels.
- Keep headings hierarchical, specific, and parallel at the same level.
- Prefer literal, unambiguous language that survives translation.
- Put conditions before instructions and expected outcomes after commands.
Validate semantics, not decoration
The documentation tests must detect more than matching headings. They verify:
- one H1 and a logical heading hierarchy;
- ordered semantic section and critical-claim markers;
- exact first-success commands and stable identifiers;
- relative links and locale-specific detail destinations;
- version badge parity with runtime metadata;
- meaningful image alt text and adjacent visual fallbacks;
- a single Mermaid lifecycle with a complete text equivalent;
- two-column table and paragraph-length budgets;
- absence of designated deep implementation detail from landing pages;
Research basis
- GitHub: About the repository README file defines the README's first-visit purpose and recommends moving long-form documentation elsewhere.
- Diátaxis separates tutorial, how-to, explanation, and reference needs.
- Microsoft: Scannable content emphasizes first-things-first structure, short paragraphs, and consistent visual entry points.
- Google developer documentation style recommends active voice, direct address, descriptive headings, accessibility, and global-audience writing.
- GitHub: Creating diagrams documents Mermaid support in Markdown.
- W3C WAI: Images tutorial requires text alternatives and complete equivalents for informative and complex visuals.