2,712 件 ・ 関連度順
概要と使いどころ
Every product will be AI-powered. The question is whether you'll build it right or ship a demo that falls apart in production. This skill covers LLM integration patterns, RAG architecture, prompt engineering that scales, AI UX that users trust, and cost optimization that doesn't bankrupt you. Use when "keywords, file_patterns, code_patterns, " mentioned.
日本語の概要は準備中です。原文の説明を表示しています。
Install a per-turn canary signal (e.g. starting every reply with the user's name and a turn counter) so silent context degradation becomes visible the moment it happens, and run a recovery protocol when the canary trips. Use when the user mentions a "canary", "context canary", or "canary check", asks to detect context rot / compaction / drift, says "you stopped using my name" or "did you lose context", asks "how degraded is your context", or wants an early-warning system for long agent sessions.
日本語の概要は準備中です。原文の説明を表示しています。
Keeps specification documents and code in agreement. Audit mode finds every divergence when an implementation is claimed complete; sync mode detects spec drift in staged changes and updates the spec after user approval, so each commit is a reconciled snapshot of spec, tests, and code. Use when an implementation is claimed complete against a spec, before committing changes that may drift from the spec, or after approving decisions that must flow back into it.
日本語の概要は準備中です。原文の説明を表示しています。
Challenges a recommendation, design, large change, or any artifact submitted for hard review (spec, proposal, policy, plan) across 5 dimensions (consistency, completeness, actionability, alignment, risk), from the standards of a named senior expert in the artifact's domain and ideally from a fresh context or a different model. Searches assuming defects exist and reports only those that survive evidence. Panel mode runs several independent lenses in parallel on a release candidate, verifies every finding against the current head, and routes confirmed ones to file owners. Use before shipping a significant recommendation, design, large branch, or multi-agent release, or when asked to tear something apart, be brutally honest, or poke holes in it.
日本語の概要は準備中です。原文の説明を表示しています。
Runs a multi-phase project from an orchestrating main thread that plans, routes and verifies but does not build. Defines a few long-lived workstreams, each owning a file area end to end in its own worktree, and routes every follow-up to the owning agent instead of spawning a one-off. Covers base-SHA recording, rebase-before-commit, stale-checkout detection, and shared-machine hygiene (ports, processes, measurement noise). Use when a project spans several phases or sessions with three or more agents working at once.
日本語の概要は準備中です。原文の説明を表示しています。
Audit experiment integrity against the evidence ledger. At L2 (repo + result files present) a fresh cross-model reviewer reads the eval code line-by-line for fake/derived ground truth, score self-normalization, phantom results (a paper number with no backing file/key), dead/uncalled metric code, verified-scope inflation, method-described ≠ method-evaluated drift, synthesized-looking results, placeholder/fake data still wired into a released result, code-output ≠ reported-number mismatch, and missing reproducibility artifacts (an empirical/agent/LLM paper shipping neither code nor the prompts/configs its results need) — every finding span-anchored to a ledger claim_id. At L0/L1 (PDF / source only) the same patterns are surfaced as info-level 'could-not-verify' signals where the ledger gives an anchor (observability_level_required:2) — NEVER a fraud verdict from a PDF. The reviewer PROPOSES findings; tools/adjudicate_findings.py computes the verdict. Detect-only. Triggers: "experiment forensics", "audit the results", "check the eval code", "实验诚实度".
日本語の概要は準備中です。原文の説明を表示しています。
Synthesize the single strongest EVIDENCE-BOUND reviewer case to reject a paper, built ONLY from the evidence ledger (claims.json) + the other auditors' confirmed findings — never free-floating LLM critique. Two fresh cross-model codex threads: an attack writes the ~200-word rejection paragraph (every accusation tagged to an existing claim_id/finding_id), a defense decomposes it and rules each point against the anchored evidence. MEMO-ONLY: emits adversarial-case-builder.memo.md (fed to the adjudicator via --memo) and carries NO verdict weight — tools/adjudicate_findings.py lists it in ZERO_WEIGHT_SKILLS and caps it at info. Honest-null allowed (the paper may survive). Run LAST. Detect-only. Adapted from ARIS kill-argument. Triggers: "adversarial case", "strongest objection", "rejection memo", "kill argument", "最强拒稿点".
日本語の概要は準備中です。原文の説明を表示しています。
Flagship intra-paper self-consistency forensics: does the paper contradict ITSELF across abstract/intro/tables/body/appendix, and does the method DESCRIBED match the method EVALUATED? Needs no external ground truth — works PDF-only (L0). Runs a deterministic arithmetic pass + a fresh cross-model semantic pass, every finding span-anchored to the evidence ledger (claims.json), reviewer≠adjudicator. Emits consistency-audit.findings.json; NEVER computes the verdict. Triggers: "consistency audit", "check the paper against itself", "self-consistency", "内部自洽".
日本語の概要は準備中です。原文の説明を表示しています。
Transparent, itemized impressions of AI-generated WRITING STYLE — the repo's ONLY non-integrity track. Two passes: a deterministic defensive-hedge density screen (tools/check_ai_style.py, AIS-DEFENSIVE-HEDGE) plus a fresh cross-model GROSS-cases-only semantic pass over the 13 AIS-* style tells (broken narrative arc, LLM phrase tics, jargon-stuffing, invented codenames, clause/formula walls, gratuitous pseudocode, bullet overuse, bold-module spam, restatement loops, focus drift, single-style figures, appendix dumping). Every finding is named, LOCATED, span-anchored to the evidence ledger (claims.json), carries not_integrity_finding:true + false_positive_risk:high + an fp_case, and gets ZERO verdict weight: the adjudicator forces it to info, excludes it from overall_verdict, and renders it in a SEPARATE report section. NOT an AI-text classifier — no scores, no "this is AI-written", no authorship probability; a paper can be CLEAN_GIVEN_EVIDENCE and still list many. Emits ai-style-impressions.findings.json; computes NO verdict; if a tell is actually substantive it routes to the integrity auditor. Triggers: "AI style", "vibe check", "writing fingerprint", "AI 文风", "vibe paper".
日本語の概要は準備中です。原文の説明を表示しています。
Build the deterministic evidence ledger (artifact_manifest.json + claims.json) that every other Anti-Autoresearch auditor reads. One pass inventories artifacts, derives the observability level (L0 PDF-only / L1 +LaTeX / L2 +repo+results) by fixed rule, and extracts span-anchored, hashed, checkable claims (numbers, comparisons, scope, method, baselines, citations, captions, table cells) into claims.json. An OPTIONAL additive cross-model pass ADDS span-anchored semantic claims — method, theorem statements with their assumptions, definitions, proof/derivation steps and equations, scope, baselines, conclusions, the motivation span, and reproducibility-artifact references (the proof, derivation, and structure anchors the family B/D/G auditors need) — it never invents a number, emits a finding, or computes a verdict. Run FIRST, before any audit skill. Triggers: "build the ledger", "extract claims", "prep for integrity audit", "evidence ledger", "建证据账本".
日本語の概要は準備中です。原文の説明を表示しています。
Audit whether a paper's EVALUATION DESIGN actually measures what it claims and whether its reporting is complete — the validity layer family D (experiment-forensics) cannot reach. Three patterns: train/test leakage means the reported score may not measure generalization (HP-EVAL-LEAKAGE — adopts the Kapoor & Narayanan 8-type / 3-category leakage taxonomy; the illegitimate-proxy / sampling-bias / pretraining-contamination subtypes hand off as needs_external_check, naming but NEVER running Oren-2023 exchangeability / Shi-2023 Min-K% / Golchin-2023 Time-Travel / BIG-bench canary); a load-bearing LLM judge is conflicted (same model/family as a compared system) or unvalidated (no human-agreement, no bias control) (HP-JUDGE-VALIDITY); a declared condition/metric is dropped or switched to favor the method, or 'best' is chosen with no held-out set (HP-SELECTIVE-REPORTING). Verdict-bearing at L0/L1 from the DESCRIBED protocol — NOT repo-gated like experiment-forensics; L2 only CONFIRMS against split/preprocessing/result files. A fresh cross-model reviewer (gpt-5.6-sol xhigh, read-only, fresh thread per pass) PROPOSES findings, each span-anchored to a ledger claim_id; tools/adjudicate_findings.py DECIDES the verdict. Leakage and under-reporting are usually HONEST methodological errors — every finding describes a discrepancy to CHECK, never an accusation. An LLM generating GROUND-TRUTH labels is HP-FAKE-GT (experiment-forensics) — routed there, not here. Emits eval-design-forensics.findings.json; computes NO verdict. Detect-only. Triggers: "eval design audit", "evaluation validity", "train/test leakage", "data leakage", "is the score measuring generalization", "LLM judge bias", "is the judge validated", "selective reporting", "cherry-picked results", "评估设计审计", "评测有效性", "数据泄漏", "训练测试集泄漏", "裁判模型有没有验证", "选择性报告".
日本語の概要は準備中です。原文の説明を表示しています。
Citation-integrity forensics: is every reference real, correctly attributed, and used in a context the cited work actually supports? Catches hallucinated references (no paper at the claimed arXiv id/DOI/venue, fabricated authors/year), metadata drift (wrong year/venue/version), and wrong-context citations (a real paper cited for a claim it never makes — or argues against). A hot zone for machine-generated papers. Decidable at L0 (text + canonical sources). Span-anchored to the evidence ledger (claims.json); the executor gathers canonical facts (DBLP / arXiv / DOI), then one FRESH cross-model thread per cited key proposes findings; reviewer != adjudicator. Emits citation-forensics.findings.json; NEVER computes the verdict. Triggers: "citation forensics", "check the references", "hallucinated citations", "wrong-context citation", "verify references", "引用核对".
日本語の概要は準備中です。原文の説明を表示しています。
AKF — The AI Native File Format workflow skill. Use this skill when the user needs The AI native file format. EXIF for AI — stamps every file with trust scores, source provenance, and compliance metadata. Embeds into 20+ formats (DOCX, PDF, images, code). EU AI Act, SOX, HIPAA auditing and the operator should preserve the upstream workflow, copied support files, and provenance before merging or handing off.
日本語の概要は準備中です。原文の説明を表示しています。
AKF — The AI Native File Format workflow skill. Use this skill when the user needs The AI native file format. EXIF for AI — stamps every file with trust scores, source provenance, and compliance metadata. Embeds into 20+ formats (DOCX, PDF, images, code). EU AI Act, SOX, HIPAA auditing and the operator should preserve the upstream workflow, copied support files, and provenance before merging or handing off.
日本語の概要は準備中です。原文の説明を表示しています。
临床诊断思维 (临床诊断思维 / 临床推理 (Clinical Diagnostic Reasoning) — 医生『怎么想病』的元学科:从症状/体征/检查到诊断结论的认知操作系统,从业者(临床医生/住院医/规培生,尤其全科/急诊/大内科等以未分化主诉为生的科室)、医学生与医学教育者、诊断安全与质量改进研究者、以及做医疗 AI 辅助诊断产品的人的视角。覆盖: (a) 第一性张力 — **直觉模式识别 (System 1: illness scripts 疾病脚本 / pattern recognition / gestalt, 『资深人一眼认出 aunt Minnie』) ⇄ 分析性推理 (System 2: hypothetico-deductive 假设演绎 / Bayesian 概率更新)** 的 dual-process 双过程理论 (Kahneman→Croskerry/Norman 谱系), 资深≠更会分析而是『脚本库更大+校准更好』; 更深层论战 — **『认知去偏可教 (Croskerry: bias awareness / cognitive forcing strategies / diagnostic timeout) ⇄ 偏倚标签是马后炮、知识结构才是主因 (Norman/Sherbino/Monteiro: debiasing 干预 transfer 证据弱, bias 是 hindsight 标签)』** — 本行最核心学术对垒; 『概率思维 (验前概率 × 似然比 → 验后概率, Pauker-Kassirer test/treatment threshold 阈值模型) ⇄ 穷尽式排查 (rule-out everything / 防御性医疗 / VOMIT)』; 『临床 gestalt ⇄ 结构化临床决策规则 (Wells/PERC/HEART)』; 『诊断简约 Occam's razor ⇄ Hickam's dictum (病人可以同时得 N 个病)』; 『床旁体格检查复兴 (Verghese Stanford 25 / McGee 循证体检) ⇄ 影像检验替代床旁』; (b) 方法论正典 — illness script theory (Schmidt/Boshuizen: enabling conditions/fault/consequences 三段结构), problem representation + semantic qualifiers 问题表征与语义限定词 (Bowen NEJM 2006, 把病人翻成 one-liner), hypothetico-deductive model (Elstein 1978《Medical Problem Solving》: 早期假设生成+定向检验), Bayesian 工具箱 (sensitivity/specificity/LR/Fagan nomogram/SnNout-SpPin), threshold model (Pauker-Kassirer NEJM 1975/1980), 认知偏倚分类学 (anchoring/premature closure/availability/confirmation/base-rate neglect/search satisficing/diagnostic momentum — Croskerry 偏倚清单), 去偏与元认知策略 (diagnostic timeout/cognitive forcing/Ely checklist/calibration), schema-based reasoning (Clinical Problem Solvers schemas: 按 pivot 症状走分支), reflective practice 结构化反思 (Mamede/Schmidt), 诊断错误科学 (NAM 2015 报告定义 / Newman-Toker Big Three / Hardeep Singh e-triggers / SAFER Dx 框架); (c) 行业结构与角色 — 医学生→实习/住院医 (晨会 morning report/查房被 pimping/汇报训练)→主治 attending→master clinician (NEJM CPC discussant/晨会大师如 Dhaliwal); 配套生态: 医学教育者 (clinical reasoning curriculum + 评估: script concordance test/key features exam/OSCE), 诊断安全研究者 (SIDM/AHRQ), 诊断辅助与 CDS 工具开发者 (DDx generator/AI); (d) 核心工作流 — 数据采集 (病史为王 + 循证体检 + 针对性检验影像) → 问题表征 (one-liner + semantic qualifiers) → 鉴别诊断生成 (schema / 解剖定位法 / VINDICATE-M 病因筛, 按『常见可能 × 致命不能漏 can't-miss』双轴排序) → 假设定向检验 (按 LR 选检查 / 阈值决策) → working diagnosis + 安全网 (red flags 交代 / test of time / test of treatment) → 反馈校准 (follow-up / M&M / diagnostic timeout); 教学工作流: 晨会渐进披露汇报 / CPC / SNAPPS / one-minute preceptor / virtual morning report; AI 增强工作流 (2023-2026: LLM 鉴别诊断头脑风暴 / OpenEvidence 检索 / ambient scribe 释放认知带宽 + 自动化偏倚 guardrails); (e) 产出物 — one-liner, problem list, prioritized DDx, assessment & plan (按问题分层), 晨会/CPC 汇报, M&M 复盘, 诊断不确定性沟通 (『最可能是 X, 但出现 Y 红旗立刻回来』); (f) 教育与评估 — script concordance test / key feature exam / EPA / 里程碑; 中国语境: 人卫《诊断学》教材 / 执业医师考试 / 规培结业临床思维考核; (g) 争议/批判 — debiasing 之争 (Croskerry vs Norman-Sherbino『knowledge is the cure』, 去偏 RCT 效果弱 / bias 标签不可证伪), 诊断错误率数字之争 (『10-15%』经典估计 / Newman-Toker 79.5 万美国年严重伤害外推方法被质疑 / 尸检符合率), dual-process 二分被批过度简化 (连续谱 / 难以实证分离), 决策规则 vs gestalt (资深 gestalt 常不输 Wells 类规则 / 算法厌恶), vignette 研究外推性 + context/case specificity (推理高度内容绑定不可通用迁移 [Norman/Eva] — 对『教推理通用课』产业的根本批判), AI 辅助诊断 2023-2026 (GPT-4 在 NEJM CPC/vignette 追平或超医生 [Kanjee 2023 JAMA / Goh 2024 JAMA Netw Open『AI alone > physician+AI』悖论] / Google AMIE / OpenEvidence 爆发 vs 自动化偏倚 / 去技能化 / 真实环境验证缺失), 防御性医疗与过度检查 (incidentaloma 瀑布 / VOMIT), 体检衰亡之争, pimping 教学法争议; (h) 流派/思想谱系 — 决策科学/Bayesian 派 (Ledley-Lusted 1959 → Elstein/Kassirer/Sox/Pauker/Eddy → Brush) vs 认知心理/双过程派 (Kahneman-Tversky → Croskerry/Graber) vs 教育认知派 (Schmidt/Boshuizen illness scripts → Norman/Eva/Mamede/Durning 情境认知) vs 诊断安全/系统派 (NAM 2015 / Newman-Toker / Hardeep Singh / Graber-SIDM: 错误=系统×认知共因) vs 床旁临床派 (Osler 传统 → Tierney aphorisms / Verghese / McGee / Dhaliwal / Saint: 病史体检为王 + 刻意练习) vs AI/计算派 (INTERNIST-1/DXplain 专家系统 → Isabel → LLM 世代 Rodman/Topol)。不含: 具体专科治疗方案与手术决策深度 (足踝外科/种植牙等另有 skill), 中医辨证 (另有 skill), EBM 文献批判性评价方法学 (相关但本 skill 聚焦诊断推理本身), 护理诊断体系 (NANDA), 影像/病理知觉型读片训练 (相邻但独立), 精神科 DSM 结构化访谈细节, 医患沟通技巧全集 (只覆盖诊断不确定性沟通), 患者自查/自我诊断指
日本語の概要は準備中です。原文の説明を表示しています。
亚马逊管理之道 (亚马逊管理之道 (The Amazon Way — Amazon 自身的领导力与运营操作系统) — 这个 master skill 蒸馏的不是『在亚马逊平台上开店卖货』(那是跨境电商运营,已由 cross-border-ecommerce 另蒸),也不是 AWS 云技术架构;而是亚马逊这家公司『怎么运营自己、怎么做决策、怎么招人、怎么创新、怎么规划』的一整套可迁移的管理机制(mechanisms)与心智操作系统。面向想在自己团队/公司引入亚马逊式机制的创始人/CEO/高管/运营负责人(COO)、产品经理与技术产品经理(PM/PMT)、组织与流程设计者(chief of staff / ops)、以及在亚马逊工作或准备面试亚马逊(著名的 LP 行为面试 + Bar Raiser)的人。覆盖: (a) 第一性张力 — 亚马逊 OS 的核心世界观分歧: **机制 > 良好意图(Good intentions don't work, mechanisms do:问题不能靠『下次更努力』修复,要装一个带检查闭环的自我强化流程) ⇄ 可控输入指标 vs 滞后输出指标(管你能控制的 input 如选品/价格/在库/配送速度,而非 revenue/股价这类滞后又不可控的 output) ⇄ 长期主义 vs 季度业绩(Day 1、willing to be misunderstood、用现金流+飞轮长期投入,宁可牺牲短期利润) ⇄ 客户执念 vs 竞争对手执念(working backwards 从客户倒推,而非从竞品或『我们能造什么』正推) ⇄ 叙事备忘录 vs PPT(6-pager + 开会前静默阅读、动手前先写 PR/FAQ;写作逼出清晰思考,禁 PowerPoint) ⇄ 单线程负责人/两个披萨团队 vs 协调税(separable、autonomous、one owner per initiative,砍掉跨团队依赖) ⇄ 高标准可学习可传递 vs 与生俱来(insist on the highest standards + Bar Raiser 招聘门,标准是 domain-specific 且能教的) ⇄ 节俭催生创造力 vs 血汗文化(frugality/约束驱动发明 —— 但 frugality + 高压 bar + URA 强制淘汰 = 文化阴暗面批判) ⇄ 两类决策(Type 1 不可逆/单向门 慎重 vs Type 2 可逆/双向门 高判断力个人快速决策,别用重流程拖慢))**; (b) 核心机制(mechanisms,本行的『工具栈』其实是一套运营机制而非软件) — Leadership Principles(14 条→2021 增至 16 条,是一套运营语言和决策仲裁器)、Working Backwards(从客户倒推:动手前先写新闻稿+常见问答 PR/FAQ)、叙事备忘录(6-pager narrative / 1-pager + silent reading 静默阅读、禁 PPT)、Weekly Business Review(WBR + metrics deck 指标看板)、input/output metrics(可控输入指标体系、DMAIC 式指标选择)、年度规划(OP1/OP2 + S-Team goals)、Bar Raiser(招聘机制,独立于用人经理的抬杠者拥有一票否决)、Two-Pizza Team / Single-Threaded Leader STL / separable teams、Correction of Errors(COE 事故复盘 + 5 Whys 五问法)、Andon Cord(安灯绳,借自丰田,一线可拉停)、Flywheel(飞轮:低价→更多客户→更多卖家→更多选品→更低成本结构→更低价)、Type 1/Type 2(单向门/双向门决策分级)、Regret Minimization Framework(遗憾最小化,贝索斯创业决策法); (c) 心智模型层 — customer obsession(客户执念是所有 LP 之首)、long-term thinking / Day 1(第二天就是停滞衰亡)、invent and simplify、ownership(主人翁,不说 that's not my job)、bias for action(可逆决策就快,speed matters)、disagree and commit(充分反对后即便不同意也全力执行)、frugality、dive deep(深入细节、不靠代理指标脱离一线)、think big、are right a lot(用强判断力和证伪自己的能力多做对的判断)、earn trust、deliver results; (d) 工作流/SOP — 用亚马逊方式发起一个新产品/业务(先写 PR/FAQ→评审→倒推着造)、如何跑一场 WBR、如何写和评审一份 6-pager(叙事流程 + 静默阅读)、Bar Raiser 招聘 loop、年度规划(OP1/OP2→S-Team goals→输入指标追踪)、事故后的 Correction of Errors; (e) 知识正典(en-primary,极其丰富且可取一手) — 贝索斯致股东信 1997-2021(1997 首封是奠基文本,每年附在信末;Day 1、遗憾最小化、long term)、《Working Backwards》Colin Bryar & Bill Carr(2021,运营机制圣经)、《The Everything Store》Brad Stone(2013,公司史)、《Amazon Unbound》Brad Stone(2021,权力与阴暗面续作)、《Invent and Wander》贝索斯文集(2020)、《The Amazon Way》/《Think Like Amazon》John Rossman(前亚马逊高管)、Amazon 官方 Leadership Principles 页(amazon.jobs,一手)、HBR『The Institutional Yes』(2007 贝索斯访谈); (f) 行业大佬/figures — Jeff Bezos(创始人,源头教义)、Andy Jassy(现任 CEO,OS 超越创始人的证据 + AWS 缔造者)、Colin Bryar(前贝索斯影子/TPM,《Working Backwards》作者)、Bill Carr(前数字媒体 VP,《Working Backwards》作者)、Jeff Wilke(前全球消费者 CEO,运营架构师)、John Rossman(前亚马逊、机制布道者/顾问)、Brad Stone(记者/编年史作者,批判性外部视角); (g) 争议/批判(必须保留,不能洗白) — 文化阴暗面(NYT 2015『Inside Amazon: Wrestling Big Ideas in a Bruising Workplace』揭露高压职场)、仓储与一线劳工争议、URA(unregretted attrition 强制淘汰)与 stack-ranking 的残酷、frugality + 高 bar = 燃尽(burnout);机制能否迁出亚马逊(cargo-culting:在初创或非科技公司照搬 PR/FAQ/输入指标/Bar Raiser 常失败);是贝索斯个人天才(personality)还是制度化机制(institution)—— OS 在贝索斯卸任后于 Jassy 治下延续,是机制派的证据; (h) 流派/思想谱系 — **信徒/机制布道派(Bryar & Carr《Working Backwards》+ 前亚马逊 operator 顾问圈,信 mechanisms 可移植、任何公司可安装)** vs **批判派(Brad Stone《Amazon Unbound》+ NYT + 劳工视角,认为 OS 与高淘汰的残酷机器不可分割,不该被浪漫化)** vs **情境/规模怀疑派(机制之所以有效是因为亚马逊特定的规模/现金流/创始人权力,照搬到小公司=cargo cult)** vs **创始人中心 vs 制度中心之争(是贝索斯个人 vs 可传承的机制)**。不含: 在亚马逊开店卖货/选品/PPC/FBA 的卖家运营(=cross-border-ecommerce)、AWS 云技术架构与运维、通用 MBA 管理理论泛泛而谈、把普通商业常识包装成『亚马逊独家秘诀』。) Master OS — automated mastery of 亚马逊管理之道 (The Amazon Way — Amazon 自身的领导力与运营操作系统) — 这个 master skill 蒸馏的不是『在亚马逊平台上开店卖货』(那是跨境电商运营,已由 cross-border-ecommerce 另蒸),也不是 AWS 云技术架构;而是亚马逊这家公司『怎么运营自己、怎么做决策、怎么招人、怎么创新、怎么规划』的一整套可迁移的管理机制(mechanisms)与心智操作系统。面向想在自己团队/公司引入亚马逊式机制的创始人/CEO/高管/运营负责人(COO)、产品经理与技术产品经理(PM/PMT)、组织与流程设计者(chief of staff / ops)、以及在亚马逊工作或准备面试亚马逊(著名的 LP 行为面试 + Bar Raiser)的人。覆盖: (a) 第一性张力 — 亚马逊 OS 的核心世界观分歧: **机制 >
日本語の概要は準備中です。原文の説明を表示しています。
Triage the deferred-work ledger for the bmad-loop orchestrator: verify every selected open entry against the actual codebase and return a machine-readable partition (bundles, already-resolved, blocked, skip, human decisions). Also migrates legacy pre-DW-format ledgers when invoked with --migrate. Automation-only — invoked by bmad-loop sweep runs, not by humans.
日本語の概要は準備中です。原文の説明を表示しています。
Runs a long swarm of hundreds to thousands of cheap headless coding workers that drive a large failing-test backlog to green. Failures are clustered by signature, each worker fixes one cluster in its own git worktree, a serial no-regression merge gate lands the fixes, the loop resumes across restarts for hours, and it federates across sibling sessions over git branches when one container's CPU is not enough. Use for "fix these thousands of failing tests with a Haiku swarm", "drive the test suite to green with hundreds of agents", "build this compiler or interpreter against a conformance suite", "launch, relaunch or check in on the swarm loop", "the swarm is stuck, idle or rejecting everything", "parallel agents fixing tests keep hitting merge conflicts and regressions", "add sibling containers to the swarm", or "compare models on the tail the swarm could not fix". Ships the harness, a per-project TOML config, a fake-worker selftest, check-in signals and the 2026-10-12 haiku-cc failure catalogue.
日本語の概要は準備中です。原文の説明を表示しています。
Routes, triages, flags and rates a piece of text with a probability for every option: which department or queue a ticket goes to, which intent a message expresses, whether a yes/no condition holds (is this tool call grounded, should the agent ask first, is the customer angry), and how severe or urgent it is on a rubric, each with a calibrated confidence. Runs Claude Haiku in the request and response shapes of OpenAI's Decisions API, for when no purpose-built decision model is at hand. Use for "route this ticket", "triage these", "classify with a confidence", "give me a probability that", "rate the severity", "score against this rubric", guardrail checks before a tool call, or "a Decisions API without OpenAI".
日本語の概要は準備中です。原文の説明を表示しています。
Preprocesses photographed sheets of many business cards — slicing each into overlapping high-resolution tiles and de-glaring them with container tooling (OpenCV/ImageMagick) — then reads every card via cheap parallel temperature-0 API calls (Haiku or Sonnet) using a distilled extraction prompt, and writes deduped contact fields to a CSV. Use when a user has photos or scans holding multiple business cards per image, mentions glare or unreadable cards, batch card transcription, contact extraction, or wants to read many cards without an expensive in-conversation pass. Triggers on 'business cards', 'card scan', 'extract contacts', 'read these cards', 'card glare', 'too many cards per photo'.
日本語の概要は準備中です。原文の説明を表示しています。
Binding-resolved Python symbol queries via pyright — every true caller (--refs), the real definition (--def), or an inferred signature (--hover) of a .py symbol, excluding the same-named false positives text search cannot tell apart. Use when a task needs ALL callers or users of a named Python symbol and grep would over-match. Python only, and only for the caller/definition question: measured 2026-07-04 on real issue-localization tasks, plain ripgrep tied or beat the semantic and indexed-regex tiers at 4-60x less wall-clock, so those tiers survive below without being a default.
日本語の概要は準備中です。原文の説明を表示しています。
DEPRECATED, renamed to orchestrating-chat-agents. Kept so existing imports of its scripts keep working; every module here loads the same module from orchestrating-chat-agents. Use orchestrating-chat-agents for parallel Claude calls over the API from claude.ai, and dispatching-subagents for fan-out in Claude Code.
日本語の概要は準備中です。原文の説明を表示しています。
Execute programs on a compiled transformer stack machine where every instruction fetch and memory read is a parabolic attention head. Demonstrates that transformer attention + FF layers can implement a working computer. Use when user mentions "llm-as-computer", "lac", "stack machine", "compiled transformer", "percepta", "parabolic attention", "execute program", or asks to run/trace programs on the transformer executor.
日本語の概要は準備中です。原文の説明を表示しています。
Symbol-level navigation of a local checkout using tree-sitter ASTs. Answers where a symbol is defined, what lines it spans, which symbols a file exposes, what a directory holds, and where a name is referenced — every answer carries exact line ranges to feed straight into a scoped read. Use for "where is X defined", "who calls X", "find the function/class named", "what's in this file", "give me the line range for", "show me the source of", "list the symbols in", or before editing a file you have not read. Each invocation auto-scans and is self-contained. Not for first-encounter repo orientation (use exploring-codebases), for what a codebase DOES rather than what it contains (featuring), for binding-resolved Python caller sets (searching-codebases), or for literal text and regex matching (plain ripgrep).
日本語の概要は準備中です。原文の説明を表示しています。