跳到主要内容

dsh-doc 精选

Create, restructure, review, audit, or migrate DeepSeek Harness Markdown documentation, package READMEs, and the documentation website using audience-first hierarchy, kind-mapped YAML metadata, bilingual line alignment, summary/contents navigation, progressive user-to-developer detail, executed-operation fact-checking, and repository validation. Use for new or revised DSH docs, docs-tree organization, documentation-quality audits and budgets, website page publishing, and bilingual documentation structure changes.

在 GitHub 打开

安装方式

cp -r .agents/skills/dsh-doc ~/.claude/skills/dsh-doc

把该 skill 目录放进客户端 skills 目录。不同客户端的目录可能不同,Claude Code 默认是 ~/.claude/skills/

整仓一次性获取:git clone --depth 1 https://github.com/deepseek-ai/deepseek-harness.git

依赖与权限

  • 无需 API Key
  • 需要联网访问外部服务
  • 依赖外部命令:git pnpm make
  • 以上为按正文与 frontmatter 自动推断的信号,仅供参考;安装前建议先阅读原文。

SKILL.md 正文

来源 deepseek-ai/deepseek-harness@d743267 · 许可 MIT · 内容未修改

DeepSeek Harness documentation

Summary

The DeepSeek Harness documentation standard: make every page searchable, newcomer-readable, and exact enough for agents and maintainers, and keep the documentation website a tested projection of repository Markdown. Apply repository AGENTS.md files and executed gates first, then this workflow for kind-mapped metadata, progressive detail, line-aligned bilingual pages, corpus audits, and website publication. Preserve one owner per fact: source, tests, generated catalogs, package READMEs, guides, Agent Notes, and scratch each keep their own kind of truth. The session-persistence-jsonl README pair is the reference example of the format.

Table of Contents

  • [Workflow](#workflow)
  • [Fact-check procedure: test, do not assume](#fact-check-procedure-test-do-not-assume)
  • [Kind system and templates](#kind-system-and-templates)
  • [Voice rules](#voice-rules)
  • [Quality criteria](#quality-criteria)
  • [Audit the corpus](#audit-the-corpus)
  • [Wordcount budgets](#wordcount-budgets)
  • [Website publication](#website-publication)
  • [Detailed references](#detailed-references)
  • [Validation](#validation)
  • [Dev Note](#dev-note)

Workflow

Follow this sequence for each requested scope. Keep the common reader path brief, but do not delete failures, ownership, limitations, or other required contracts merely to reduce words.

  1. Read root and more-specific AGENTS.md, [the documentation standard](../../../docs/AGENTS.md), the target page, its source/tests, navigation owner, and bilingual record.
  2. Classify the page by one primary job and reader: product quick start, user task guide, contributor tutorial, architecture overview, package/subsystem reference, generated reference, agent instruction, decision record, or scratch.
  3. Place the page at its nearest owner. Keep package contracts beside package code; use docs/ for cross-package learning, user, developer, architecture, discussion, and expiring scratch material.
  4. Define the reader's starting state, observable outcome, likely failure, recovery path, and next useful depth before writing details.
  5. Add or revise YAML metadata — assign the kind that maps to the template for this document's job — then write Summary, Table of Contents, user-facing content, developer-facing content, optional Further Exploration, and final Dev Note in that order where the document type permits.
  6. Update the bilingual counterpart in the same pass. Keep headings, lists, tables, code, links, frontmatter layout, and physical line count aligned.
  7. Verify every claim against code, tests, generators, package metadata, or a current decision owner — and run the operations the page instructs, per the fact-check procedure below. Update the owner before any derivative artifact.
  8. Run focused checks, then pnpm run test:docs, pnpm run doc-sync, pnpm run lint, and git diff --check; re-read the complete diff for correctness and then for brevity and repository fit.

Fact-check procedure: test, do not assume

Documentation states how the product behaves today, and the only admissible evidence for an operation claim is having run it. This procedure is mandatory for every new document and every new paragraph that claims an operation, command, default, error, or platform difference.

  1. Classify the subject before writing install guidance. Read the facts, never the folder name: package.json for a dsh.bundle.patch declaration, and the entry file for the plugin shape (apply export or a default service export is a plugin; a plain module API is a library). A bundle installs with dsh plugin --profile <name> add <package> and is the only package shape for which that command activates a profile layer; a plugin mounts as a cordis.yml row; a library is a dependency with no install path of its own. Packages with special status (libraries, bundles) get their own README template — never a plugin README with install guidance that does not apply.
  2. Run every claimed operation against the current checkout. Execute each CLI command, config snippet, and profile or patch example exactly as the document will show it; write down only what you observed, including the exact output, warnings, and failure modes. If a claim depends on a key or a network you do not have, say so and name the verification owner instead of asserting the behavior.
  3. Delete what you could not reproduce. Never carry a command, field, default value, or behavior from memory, analogy, or a neighboring package's README. When a claim fails to reproduce, fix the claim — not the test.
  4. Check old docs against latest master. Before revising pre-existing pages, git fetch origin and compare the section against origin/master; the pairing sidecar recovers the last-confirmed text of either side. A stale statement on master is still wrong: correct it against the code, not against the old prose.
  5. Re-record the pair after every edit. Each paired edit re-runs pnpm run verify-translation-pairing --write <pair> so the sidecar tracks the confirmed pair.

Kind system and templates

The kind frontmatter field selects exactly one README template. Every kind in [the metadata reference](references/metadata-links-i18n.md#the-kind-system) maps to one template file in [templates/](templates/), and every template backs exactly one kind; the documentation check derives the expected kind from the same mechanical facts.

  • package-group → [templates/package-group.md](templates/package-group.md): group maps (packages/README.md, packages/<group>/README.md) — orient the family, map its direct packages, link package-owned details.
  • package-reference → [templates/package-reference.md](templates/package-reference.md): a Cordis plugin or service package — mount configuration, the config table, folded implementation, Model Experience and Known Limitations in the gate-owned forms.
  • package-library → [templates/package-library.md](templates/package-library.md): a package with no plugin surface — consumer entry points, no profile-install path, no mount configuration.
  • package-bundle → [templates/package-bundle.md](templates/package-bundle.md): a package declaring dsh.bundle.patch — the verified dsh plugin install path, layer semantics, patch document.
  • persistence-change → [templates/persistence-change.md](templates/persistence-change.md): a dated record in docs/persistence-changes/ — acknowledge a detected type transition with per-root predecessors and generated after schemas.
  • persistence-release → [templates/persistence-release.md](templates/persistence-release.md): a pinned tag in docs/persistence-changes/releases/ — compare reconstructed historical types without claiming compatibility acknowledgement.
  • persistence-format → [templates/persistence-format.md](templates/persistence-format.md): a historical Session format checkpoint in docs/persistence-changes/historical-formats/, with complete schemas; the current writer uses the existing catalog.
  • upgrade-guide → [templates/upgrade-guide.md](templates/upgrade-guide.md): docs/upgrade-guide/v<version>/<item>/guide.md and its Chinese sibling — migration steps for one externally perceptible break, owned by [dsh-create-upgrade-guide](../dsh-create-upgrade-guide/SKILL.md).

Open the template before writing and follow its skeleton and rules; it states what the kind is, how the page is structured, and the fact checks each section owes. Add a new kind only together with a distinct template file, a documented repository position or declared owner, and a focused check that maps documents to it.

Voice rules

These rules decide what a section may say. They apply to every authored human-facing page, and to package READMEs with particular force.

  • Summary says what the subject does. The opening Summary and the user-facing sections describe what a user or agent can DO with the subject — outcomes, benefits, when to choose it, main cost — never its role, type, or internal identity. In a package Summary, “what it is” means only its reader-visible capability, not its Cordis role, registrations, or internal components. Omit source identifiers unless the reader directly uses them in configuration, a command, or a public API. "The seam registers ctx.x and appends x/event records" is identity narration; "you can save a note per message and it survives restarts" is what it does.
  • Developer sections explain, never enumerate. Folded implementation content covers the overall design concept, architecture, and hand-waving dataflow — enough to understand how the package works — and links code for exact detail. No full API catalogs, exhaustive column lists, event-payload enumerations, or JSDoc restatement inside the folds.
  • Dev Note is the only slop zone. Partial ideas, scratches, undecided directions, measured artifacts, and working hypotheses live only in the final Dev Note, marked explicitly non-authoritative. Every other section is polished, current-state prose.
  • Current state only. Ordinary documentation describes the current codebase. Dedicated persistence-change, persistence-release, and persistence-format records retain historical type evidence; release comparisons and format checkpoints describe observed history without asserting compatibility. An upgrade-guide describes the change from its version directory's release to the next one. Agent Notes and postmortems retain their own historical scope.
  • Use controlled technical English. Give each sentence an explicit actor and one main action when ambiguity can change behavior. Reuse one term per concept, prefer direct verbs, split stacked instructions and conditions, and preserve modality and exceptions. Apply the non-certified, ASD-STE100-inspired discipline in [the page-style reference](references/style.md#controlled-technical-english). Do not force a shorter sentence when precision would fall.

Quality criteria

Use these definitions in review. Each section opens with a short orienting paragraph before subsections or exhaustive detail.

  • Brief: the common path contains only facts needed for its outcome; exhaustive truth remains one direct link or detail layer away.
  • Intuitive: prerequisites precede dependent concepts, one next action is obvious, and headings use terms readers search for.
  • Friendly: readers can recognize success, understand risk before acting, recover from likely failure, and choose whether to continue deeper.
  • Accurate: each durable claim has one owner and a verification path proportionate to its risk.
  • Agent-readable: metadata, stable headings, anchors, terminology, ownership, and current/proposed status support targeted retrieval without loading the corpus.
  • Newcomer-complete: a professional engineer with no repository context can reconstruct the relevant architecture or feature through three to five linked pages.

Do not apply a universal word limit to exhaustive references. Measure entry-path length, unrelated material scanned for one lookup, largest section, heading count, and page size; split by an existing domain owner when retrieval cost is high.

Audit the corpus

Read, do not re-summarize, the owning contracts: [docs/AGENTS.md](../../../docs/AGENTS.md) for hierarchy, tutorial/reference forms, taxonomy, budgets, and the slop checklist; [.agents/notes/README.md](../../notes/README.md) for Agent Note lifecycle; [docs/i18n/README.md](../../../docs/i18n/README.md) for the bilingual pairing rules; and [root AGENTS.md](../../../AGENTS.md) for standing orders. Exclude .agents/notes/archived/ from audits and edits — archived notes are frozen history.

Apply the standard's authoring order to every human-facing document in scope (not to Agent Notes): locate the document and state its own subject; set the permitted detail level and move deeper explanations to owning descendants with links; classify tutorial or reference from intended use, not path; for a tutorial, order concepts by prerequisite and difficulty; split substantial mixed forms. Then check placement constraints: paired docs cost a counterpart update and a --write re-record on every edit; generated catalogs are never hand-edited; a move is atomic with every inbound link repaired in the same change.

After the structural pass, hunt the slop checklist with the cheapest probes first. Use [dsh-trim-cot-leakage](../dsh-trim-cot-leakage/SKILL.md) for reasoning-transcript leakage, grep distinctive phrases to find duplicated rules, replace hand-written catalogs and status inventories with their authoritative owners, and remove migration plans and future-tense spec language from implemented Agent Notes. Measure outliers with pnpm run verify-doc-budgets --list and a word-count scan; if removing prose changes a promised behavior rather than its explanation, propose the behavior change first (follow [dsh-find-simplifications](../dsh-find-simplifications/SKILL.md)). Keep every load-bearing rule, preferably as one to three lines plus a link to its rationale; do not create a new explanation merely to relocate disposable reasoning.

Wordcount budgets

pnpm run verify-doc-budgets compares standing documents against ceilings in [scripts/doc-budgets.manifest.json](../../../scripts/doc-budgets.manifest.json); a red gate follows the ordered relocate-condense-raise policy in [docs/AGENTS.md](../../../docs/AGENTS.md#wordcount-budgets). Ceilings are guardrails, not reduction targets: at or below target, retain at least 5% headroom; raise a ceiling only when the words need the space, and justify the manifest diff in the PR.

Website publication

The website is a tested projection, never a second copy: [website/docs.ts](../../../website/docs.ts) is the explicit public allowlist mapping canonical docs/ sources into route trees, [scripts/project-doc-site.ts](../../../scripts/project-doc-site.ts) rewrites them into the disposable website/.generated/ tree, and VitePress builds that tree. Repository Markdown stays the only editable content source; translations stay sibling pairs (foo.md, foo.zh.md, foo.i18n.yaml), never locale directories. Edit an already published page in its canonical source only; add one manifest entry for a new page; update source, manifest entry, and inbound links atomically for a move or removal; never edit website/.generated/, website/.cache/, or website/.dist/. Set every DocsPage field deliberately and honor the projector's link rules; see [references/website-sync.md](references/website-sync.md) for the fields, sidebar collections, and preview commands. Synchronizing content into the build does not publish it: deployment stays a separate, explicitly requested step.

Detailed references

Load only the reference needed for the task. Each reference links directly from this file so the skill has no deep reference chain.

  • [Metadata, links, and bilingual pairs](references/metadata-links-i18n.md): README frontmatter, the kind system and its derivation, description semantics, repository paths, line alignment, and the sidecar record.
  • [Page structure and hierarchy](references/structure-hierarchy.md): mandatory section order, section summaries, user-to-developer progression, docs tree placement, small rule files, Further Exploration, and Dev Note ownership.
  • [Page style](references/style.md): short Summary, ----- section separators, foldable content sections, and emphasis discipline.
  • [Review criteria](references/review.md): newcomer test, evidence checks, package README review, the reference example, and verification commands.
  • [Website publication](references/website-sync.md): manifest fields, projector link rules, preview and validation, and deployment separation.

The templates in [templates/](templates/) provide one working skeleton per kind; open the one your document's kind names before writing.

Use [dsh-prose-standard](../dsh-prose-standard/SKILL.md) for sentence-level contract coverage and editorial judgment. The session-persistence-jsonl README pair ([English](../../../packages/session/session-persistence-jsonl/README.md), [Chinese](../../../packages/session/session-persistence-jsonl/README.zh.md)) is the reference example: searchable YAML, Summary and Table of Contents, user-to-developer progression with a folded developer section, Further Exploration, canonical Model Experience and Known Limitations sections, and a final Dev Note.

Validation

Validate the affected format, not merely Markdown syntax. A strong promise needs a focused valid fixture and an invalid fixture that proves the top-level gate can fail.

  • README metadata: parse YAML, map kind to its template and document standard, reject name, audience, ungoverned tags, and README-local i18n metadata, and reject missing or advertisement-style descriptions.
  • Bilingual pages: verify structure, exact line count, terminology, link parity, and the sidecar record.
  • Tutorials: exercise the documented entry path or name an explicit manual verification owner.
  • Generated references: run the deterministic freshness check and report retrieval-size measures.
  • Package READMEs: run the Summary gate, which limits each English Summary to 100 wc -w-style words and directs failures back to this skill and the kind template; run model-experience and limitation checks, then package-focused tests when behavior claims changed; re-run every command the README instructs before merging a claim about it.
  • Skills: run the repository's skill-invocation metadata check.

Run pnpm run test:docs for the quick comprehensive documentation checks (pairing, wrap, links, README gates, budgets, skill metadata, Agent Note gates) before the full pnpm run doc-sync.

Dev Note

None.

质量与来源

所属仓库
deepseek-ai/deepseek-harness
仓库星标
25 万
最近更新
3 天前
采集时间
今天
commit
d74326738864
内容指纹
sha256:a0f5a09bf759a75
文件路径
.agents/skills/dsh-doc/SKILL.md

同仓库其它 Skill

查看全部 28 个
agent-experience
Use when writing or changing a model-facing tool definition (name, description, parameter schema, or its system-prompt section), and when designing skills, context loading, or multi-step workflows, to make information discoverable and use context efficiently.
精选 deepseek-ai/deepseek-harness 25 万 原文
Claude CodeCodex CLIMCP 客户端
dsh-archive-agent-notes
Use when adding, auditing, pruning, archiving, restoring, or reviewing Agent Notes in deepseek-harness; checks every new note for superseded active records, deletes small UI and purely mechanical records, classifies other implemented notes by future decision value, deletes rejected notes that no longer prevent a tempting fallacy, and applies the frozen archived/{kind} triplet and manifest rules.
精选 deepseek-ai/deepseek-harness 25 万 原文
Claude CodeCodex CLIMCP 客户端
dsh-ci-test-reliability
Design, review, and diagnose DeepSeek Harness tests and fixtures that can fail nondeterministically under CI concurrency, shared host resources, clocks, process-global state, subprocesses, network listeners, or asynchronous teardown. Use when adding or changing tests with those risks, investigating flaky CI, or reviewing test isolation; use dsh-pre-push-checks separately to select outgoing commands.
精选 deepseek-ai/deepseek-harness 25 万 原文
Claude CodeCodex CLIMCP 客户端
dsh-client-ui-ux
Design and review DeepSeek Harness client UI changes — visual token discipline, reuse-before-adding, feedback surfaces (toast vs in-place notice vs empty state), overlay and menu safety, platform window adaptation, loading states, and error copy placement. Use when adding or changing product-user-visible GUI behavior in packages/client, or when reviewing such a PR.
精选 deepseek-ai/deepseek-harness 25 万 原文
Claude CodeCodex CLIMCP 客户端
dsh-code-review
Use when reviewing a pull request in the deepseek-harness repo — orients the reviewer to this codebase's standards (AGENTS.md conventions, defensive patterns, ADRs, quality gates) and the review-specific checks that code alone can't show
精选 deepseek-ai/deepseek-harness 25 万 原文
Claude CodeCodex CLIMCP 客户端
dsh-create-upgrade-guide
Use when a deepseek-harness change breaks an externally perceptible surface (CLI, profiles, cordis.yml or settings keys, persisted user data, SDK or wire APIs, published package entry points), or when changing a surface that an unreleased upgrade guide already covers, to write or update docs/upgrade-guide/v<version>/<item>/guide.md.
精选 deepseek-ai/deepseek-harness 25 万 原文
Claude CodeCodex CLIMCP 客户端

相似 Skill

更多「写作」
interview-prep
Prepares the user for a specific interview from their real experience. Builds a story bank in Situation/Task/Action/Result/Reflection form mapped to the competencies the role and company test (such as Amazon's Leadership Principles), writes a cited company and role brief from public sources, predicts questions per round, drafts opener, why-us, weakness and salary answers, prepares questions to ask and a logistics checklist, and runs a post-interview debrief. Use when the user says "help me prepare for my interview", "I have an interview tomorrow" or "in 2 hours", "build my story bank", "STAR stories", "interviewing as a career changer", "what will they ask me", "how do I answer tell me about yourself", "questions to ask the interviewer", "research the company" or "debrief my interview". Uses Reactive Resume MCP tools when connected. For scored role-play practice, mock-interview; for thank-you notes and tracking, job-application-manager; for a salary number, offers or negotiation, offer-negotiation.
精选 reactive-resume/reactive-resume 4.4 万 原文
Claude CodeCodex CLIGemini CLI
job-application-manager
Runs the job-search pipeline. Adds jobs from a link or pasted posting, keeps stages and dates accurate, sets follow-ups, runs a weekly review (stale applications, next actions, funnel stats), records the exact resume version sent, and drafts networking, LinkedIn, referral, recruiter-reply, follow-up, thank-you, status-check, expedite and withdrawal messages without sending them. Use when the user says "add this job to my tracker", "I applied", "log my interview", "I got rejected", "I was ghosted", "log my offer", "track my applications", "do my weekly review", "what should I follow up on", "write a thank-you note", "message this recruiter", "ask for a referral", "withdraw my application", or wants a job-search spreadsheet. Works with a Markdown/CSV tracker or Reactive Resume MCP tools. Not for resume work (resume-tailor, resume-bullet-writer, resume-content-guide), interview prep or practice (interview-prep, mock-interview), or judging, negotiating, accepting or declining an offer (offer-negotiation).
精选 reactive-resume/reactive-resume 4.4 万 原文
Claude CodeCodex CLIGemini CLI
mock-interview
Runs a mock interview for a specific job description. Plays the interviewer in a chosen round (recruiter screen, hiring manager, behavioral, coding, system design, case, product sense, panel), persona and difficulty; asks one question at a time, probes vague answers, and gives quote-backed rubric scores after each answer (coach mode) or at the end (realistic). Model answers use only the user's real experience. Use when the user says "mock interview me", "interview me for this job", "practice interview", "pretend you're the interviewer", "grill me for this role", "run a behavioral round", "quick recruiter screen", "hiring-manager mock", "system design mock" or "case interview practice". With Reactive Resume MCP it reads the application and stories and logs the session. For a story bank, company brief, questions to ask or a real-interview debrief, interview-prep; for scheduling or thank-you notes, job-application-manager; for fixing the resume, resume-tailor; for negotiation practice, offer-negotiation.
精选 reactive-resume/reactive-resume 4.4 万 原文
Claude CodeCodex CLIGemini CLI
offer-negotiation
Evaluates and negotiates job offers. Decodes base, bonus, equity, sign-on, level and terms with unknowns kept unknown and each offer in its own currency, benchmarks against checkable pay data, finds true leverage, drafts one counter as email and call script, role-plays the recruiter, and handles the salary-expectations number, deadlines and extensions, "best and final", competing offers, counteroffers and written acceptance or decline. Country-aware (India CTC, 13th-month pay, public-sector scales). Use when the user says "I got an offer of...", "is this offer good", "should I negotiate", "how do I counter", "compare my offers", "what is my equity worth", "they asked my salary expectations", "they want an answer by Friday", "can I ask for more time", "should I accept" or "practice the negotiation call". Uses Reactive Resume MCP tools when connected. For logging the offer, its deadline and withdrawal notes, job-application-manager; for other interview answers, interview-prep.
精选 reactive-resume/reactive-resume 4.4 万 原文
Claude CodeCodex CLIGemini CLI
resume-builder
Builds a complete resume from scratch as Reactive Resume JSON that conforms to the schema and imports at https://rxresu.me. Asks for each missing detail instead of inventing it, then produces valid JSON (UUIDs, HTML descriptions, template and layout). Use when the user wants to create, build or generate a new resume, turn their work history into Reactive Resume JSON, or asks about the Reactive Resume schema, sections, templates or layout. For what belongs on a resume or a resume audit use resume-content-guide; for rewriting bullets, resume-bullet-writer; for fitting one job posting, resume-tailor; for tracking applications, job-application-manager; for interviews, interview-prep or mock-interview.
精选 reactive-resume/reactive-resume 4.4 万 原文
Claude CodeCodex CLIGemini CLI
resume-bullet-writer
Turns resume duties into truthful impact bullets. Questions the user role by role to dig out achievements, asks for real numbers or honest ranges (never invents them), and writes concise bullets in XYZ, CAR, scope or qualitative form. Use when the user says "rewrite my bullets", "make this bullet stronger", "quantify my resume", "turn my duties into achievements", "improve my experience section", "my resume sounds boring", "I don't have any numbers", pastes resume lines for feedback, or works in a low-number job (nursing, teaching, care, trades, admin, creative). With the Reactive Resume MCP server connected, it edits experience, project, volunteer and education descriptions with apply_resume_patch after the user approves. Use resume-content-guide instead for which sections or details belong and for summaries or headlines, resume-tailor to fit one job posting, interview-prep or mock-interview for interview stories and practice, and resume-builder to build a resume from scratch.
精选 reactive-resume/reactive-resume 4.4 万 原文
Claude CodeCodex CLIGemini CLI