跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
开发与测试

documentation

Write or revise Labtasker's README and user documentation, especially product positioning, tutorials, guides, examples, case studies, and navigation. Preserve the project's direct, plain-English style for ML researchers, agent-friendly workflow, and contract accuracy. Do not use for code-only work; pair observable behavior changes with the public-contract-change skill.

luocfprime · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add luocfprime/labtasker --skill documentation

技能说明


name: documentation description: Write or revise Labtasker's README and user documentation, especially product positioning, tutorials, guides, examples, case studies, and navigation. Preserve the project's direct, plain-English style for ML researchers, agent-friendly workflow, and contract accuracy. Do not use for code-only work; pair observable behavior changes with the public-contract-change skill.

Write Labtasker documentation

Write for an ML researcher or engineer who has independent inference, evaluation, or experiment jobs but has never used Labtasker. The reader may not know task-queue terminology. Help them understand what Labtasker is within one sentence, recognize its main benefits within one short section, and decide whether to use it before teaching the internal model.

Put definition and value first

The README and documentation homepage should follow this reader journey:

  1. One direct sentence defining Labtasker.
  2. Prominent links to the documentation, llms.txt, and source code.
  3. A short paragraph explaining what Labtasker adds to existing ML code.
  4. Three to five consolidated key features.
  5. Installation and a short representative example.
  6. Separate When to use Labtasker and When NOT to use Labtasker sections.
  7. Navigation to tutorials, concepts, guides, and reference pages.

Do not begin with a long hypothetical workload, implementation details, v1 history, or internal terminology. Put fuller motivation and v1-to-v2 rationale in docs/why-labtasker.md.

Keep the README and documentation homepage aligned on the definition, features, and product boundary. The README may contain a compact example; the homepage should focus on navigation after establishing the product.

Write features as claims with evidence

Use the SQLModel-style pattern **Short benefit:** Concrete explanation. A positive adjective such as effortless is useful when the following sentence immediately demonstrates why it is true.

Consolidate related capabilities instead of listing every mechanism separately. The current product story fits four groups:

  • effortless and flexible parallelism;
  • resumable and failure-resistant experiments, supported by tested lifecycle behavior;
  • structured Task records for inspection; and
  • easy adoption and use, including end-to-end operation by agents.

Prefer user-visible capabilities such as automatic retry, dynamic priority, cancellation, recorded results, and agent operation. Do not lead with SQLite, local daemon startup, dependency separation, leases, or run_id fencing. Those details belong where readers configure or verify the behavior.

Do not narrow the entire product to one incidental condition such as uneven runtimes or one benchmark type. Concrete scenarios are useful examples, not the definition of Labtasker's scope.

Use direct, natural language

Use short sentences that say what the reader can do. Prefer familiar ML words such as jobs, cases, GPUs, failures, results, and restart before introducing Task, Worker, Queue, route, attempt, lease, or run_id.

Avoid infrastructure metaphors and compressed abstractions on landing pages:

  • Write “resume after an interruption without rerunning completed jobs,” not “jobs survive interruption.”
  • Write “when a Worker stops responding,” not “abandoned Tasks.”
  • Write “matching Task,” not “eligible Task,” until matching rules are taught.
  • Write “results from old runs,” not “stale results,” until run_id is taught.
  • Write “a simple loop can be sufficient,” not “a simple loop is better.”

Avoid vague words such as consistent, robust, advanced, powerful, or easy unless the same item gives a concrete reason. Do not use em dashes; split the thought into shorter sentences or use commas, parentheses, or a colon.

Prefer plain English over idioms, metaphors, and colloquial shortcuts. Write “this example uses addition to demonstrate the workflow,” not “addition stands in for inference.” Write “Workers take Tasks from the same Queue,” not “Workers draw from the backlog.” A reader should not need to interpret a turn of phrase before understanding the product behavior.

Use Labtasker terminology consistently and capitalize public concepts: Task, Queue, Worker, Client, and Server. Use lowercase route because it is a label, not a resource record.

Wording blacklist

Never use these phrases or close variants in user-facing documentation:

  • provide the compute, supplies the compute, or compute you control.

These phrases sound like infrastructure-provider language and obscure the actual product boundary. When that boundary matters, state the concrete fact, such as “Labtasker does not allocate GPUs or start machines” or “You start the Worker processes.” Do not replace a blacklisted phrase with a synonym that has the same problem.

Before handing off a documentation change, search every changed user-facing file for the listed phrases and read the changed prose once for close variants. Rewrite every match in terms of the specific action or boundary that matters.

Treat explicit user feedback as an input to this blacklist. When a user says they dislike a type of wording and that wording is common enough to recur in Labtasker documentation, update the blacklist as part of the same task. Record the general pattern rather than only the sentence that triggered the feedback, explain why it should be avoided, and give concrete rewriting guidance. Follow one-off or context-specific wording preferences in the current edit, but do not turn them into a repository-wide rule unless they describe a recurring pattern.

Explain why before how

Navigation and page order should let readers decide whether Labtasker applies before asking them to learn its model. Put Why Labtasker? before How Labtasker works.

Use one primary Diataxis type per page:

  • tutorials lead the reader through a complete successful workflow;
  • guides solve one concrete task;
  • concepts explain
阅读完整来源 ↗