跳到主要内容

dsh-trim-cot-leakage 精选

Use when auditing or fixing prose that reads like a leaked reasoning transcript — dead design-session citations such as (decision N), audit item codes, or §N of uncommitted drafts; change narration such as "used to", "no longer", "this cut"; stack or review vantage ("a later PR in this stack", "rejected in review"); reviewer-addressed justifications; control-flow narration; or hedged planning residue in comments, JSDoc, docs, or Agent Notes.

在 GitHub 打开

安装方式

cp -r .agents/skills/dsh-trim-cot-leakage ~/.claude/skills/dsh-trim-cot-leakage

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
  • 无需联网
  • 无外部命令依赖
  • 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。
触发场景
auditing

SKILL.md 正文

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

Trimming Chain-of-Thought Leakage

Chain-of-thought leakage is prose whose vantage is the authoring session rather than the repository: it cites artifacts only that session could see, narrates the change instead of the state, or argues with a reviewer who has left. The fix is never deletion alone when a passage carries factual clauses — restate each so it stands at HEAD, then delete the transcript around it; a passage carrying none (an audit code, control-flow narration) is deleted outright. REQUIRED BACKGROUND: [dsh-prose-standard](../dsh-prose-standard/SKILL.md) owns the complete-proposition rule this skill applies; the [archived committed-artifact-citations note](../../notes/archived/process/2026-08-09-committed-artifact-citations.md) records the citation rule's historical rationale. It is guidance, not a script.

The one test

For every suspect passage ask: could a reader at HEAD, with no access to any session transcript, PR thread, or uncommitted draft, resolve every reference and verify every claim? If no, restate the surviving facts from the repository's vantage and delete the rest. If yes, it is not leakage, however historical it sounds — but resolvability only clears this skill's bar: on current-state surfaces (READMEs, docs, JSDoc) a resolvable change story is still change narration, and class 3 routes it to its sanctioned home.

Taxonomy

  1. Dead design-session citations — (decision 7), (audit C2), design §4.7, plan §1.4, phase labels (T4, W3, P-I), "the design ledger", "(B ruling)". If the decision has a committed owner, cite it by name and path; otherwise delete the citation and restate its factual clause to stand alone.
  2. Stack and PR vantage — "a later PR in this stack", "this PR adds", "the previous commit". State the shipped mechanism or the extension point; deferred work moves to a TODO marker or an issue reference.
  3. Change narration and version stamps — "used to", "no longer", "the old X", and indexical stamps ("v1", "this cut", "today", "now" contrasting with a past state). State the present behavior; a fixed regression becomes a present-tense counterfactual ("without X, Y happens"), never repo history ("used to Y").
  4. Review choreography — "Rejected in review:", "the reviewer confirmed", draft ordinals ("v5 of this note"), round attributions. Keep the surviving decision and rationale as plain fact; delete who said it when.
  5. Reviewer-addressed justification — "the cast is safe — it simply…", "this is correct because…". A comment arguing its own correctness addresses a reviewer, not a maintainer. State the invariant that makes the code safe, or delete the comment if the code shows it.
  6. Restatement and derivation transcripts — control-flow narration ("first we X, then we Y"), test walkthroughs, proofs of obvious branches. Delete; keep only a non-obvious contract or invariant.
  7. Hedges and planning residue — "probably fine for now", "should be enough", deferrals with no marker. Promote to TODO/FIXME or restate as the actual bound; delete the hedge.
  8. Authoring-language slips — untranslated working-language fragments (端, 设计稿, ---- 私有 ---- separators) in prose whose language is otherwise English, or the reverse in a zh counterpart. Translate or delete.

What is not leakage

Unaided citation passes fail in both directions by deleting durable references and keeping dead ones. Apply these keep rules as written; [examples](references/examples.md) calibrates each:

  • Issue references — #1470, TODO(name):, "issue #N owns the follow-up" resolve at HEAD; keep them on any surface, including READMEs. Do not relocate them to Agent Notes.
  • Merged-PR and issue citations inside Agent Notes and postmortems — sanctioned evidence per the [documentation standard](../../../docs/AGENTS.md)'s change-story routing.
  • Suppression justifications — oxlint-disable … -- reason, coverage-ignore reasons, empty-catch explanations are required prose; fix a false reason, never delete it.
  • Counterfactual-present regression pins — "without X, Y happens", "a naive X would…".
  • Measured bounds — "(measured: 512 nests ≈ 0.15s)" calibrating a constant; the evidence word "measured" is load-bearing.
  • Runtime old/new states — "the old connection drains before the new one accepts" is runtime lifecycle, not change history.
  • Historical stage names inside a note's change-story sections — "the first cut shipped X" is current-state-safe there; indexical stamps ("this cut") stay banned everywhere.
  • External references that resolve outside the repo by design — standards sections (RFC 9110 §10.1.5), Figma frame names; the §-ban covers uncommitted internal drafts, not external standards or committed docs that own their §-numbering.
  • Project voice and genre forms — "we" as project voice; a note's Alternatives-considered section.

Workflow

  1. Scope and exclusions per [dsh-prose-standard](../dsh-prose-standard/SKILL.md): require an explicit scope; never touch vendor/ or .agents/notes/archived/. Recorded fixtures and snapshots are derivatives, not prose targets: change the owning source or scenario and regenerate them only when an authorized behavior change requires new evidence.
  2. Audit read-only first: run the [recall batteries](references/recall-batteries.md) (with --hidden so .agents/ is searched), calibrating each probe against a known positive and a near-miss negative before trusting its output, then judge every hit semantically. The batteries are probes, not the definition — each review round of the original purge found cases the batteries missed, so also read the densest prose in scope (module JSDoc, READMEs, Agent Notes) without a pattern in hand.
  3. Fix owner-first per surface: generated catalogs → trace every consumer, fix the source JSDoc or generator template, then regenerate all derivatives; type-equivalence fences → fix the source JSDoc, then re-paste both bilingual pages (verify-type-equiv pins them); bilingual prose → update the counterpart minimally and re-record it through the [lightweight routine](../../../docs/AGENTS.md#writing-rules); bilingual fences → copy the corrected verbatim block byte-for-byte into both sides per [dsh-doc](../dsh-doc/SKILL.md), then re-record the pair; model- or user-visible strings → route through [dsh-prose-standard](../dsh-prose-standard/SKILL.md) and change only with owning behavior evidence, otherwise leave unchanged and report the deferral.
  4. Before deleting anything, enumerate the passage's propositions (prose-standard) and check the [overcorrection traps](references/examples.md#overcorrection-traps): trims that flip an obligation into an endorsement, promote a hypothetical to a shipped feature, delete a true fact, or drop the evidence source.
  5. Verify: re-run the batteries expecting only sanctioned keeps, this skill's own directory, and the owning note's quoted evidence; confirm every remaining citation resolves at HEAD; run the gates for touched surfaces (doc-sync for docs, verify-type-equiv, verify-translation-pairing).

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
sha256:977eeb1dfe694e1
文件路径
.agents/skills/dsh-trim-cot-leakage/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