Latest Agent Skills
Everything added to the registry recently, newest first.
- a-horde-o-bees BundleExport Diff
Use when a change set should be reviewed visually outside the terminal — two versions of a file or directory tree, or an existing unified diff from any tool, rendered to a self-contained side-by-side HTML page set (no git required; e.g. before/after snapshots of docs, skills, or generated trees). Sets itself apart from line-based diff viewers by attributing content that moved, merged into another line, or split apart — even across line breaks — with word-level marks on only the real edits, wrapped prose lines, and a dropdown viewer for large multi-directory diffs.
- newsprint-ai BundleImprovement Loop
Run an improvement loop on your own work for the user. Look back at what you did and what went wrong, check what changed in the tools you use, and suggest up to five improvements for the user to accept or reject. The first run audits the user's workflow and AI use, with read-only access they grant source by source. Then one round runs each week, and each round checks whether the last round's changes helped. Changes nothing without the user's yes. Use for "run the improvement loop", "how can you get better", "study my week", "audit my work", "where does my time go", or "what changed in my tools".
- newsprint-ai BundleWrite Release
Write or fix a press release or announcement. Gets the facts first, matches the company's past releases, drafts quotes for approval, and never invents a figure. Also explains how to send it.
- wangyuanchuan2022 SkillParallel Execution Optimizer
触发:并行加速、多 agent 并发、批量工具调用、并行验证、怎么更快做完|English: parallelize, concurrent agents, batched tool calls, speed up|Parallelize without losing correctness: agents, batched calls, worktrees, verification lanes.
- wangyuanchuan2022 SkillAgent Harness Construction
触发:agent 工具面设计、动作空间设计、observation 格式、工具定义优化|English: agent tool design, action space, observation format|Design and optimize agent harnesses for higher completion rates.
- wangyuanchuan2022 BundleOperator Approval Loop
触发:外发消息审批、代理发信前人工确认、审批账、外发草稿审批|English: outbound message approval, operator approval, approval ledger|Operator approval contract: hashed drafts, epoch-keyed decisions, delivery claims/receipts, baseline gate.
- wangyuanchuan2022 SkillRequirement Interview
触发:需求访谈、帮我问清楚、别假设、我有个模糊想法、需求不明确、先访谈再动手|English: requirements interview, clarify the ask, don't assume, vague idea|需求访谈:苏格拉底式提问+数学化含糊度门控——Round 0 锁组件拓扑、一次一问瞄准最弱组件×最弱维度;含糊度≤阈值且用户批准前禁改代码;spec 落盘标 pending approval。
- wangyuanchuan2022 SkillAgent Workflow Router
触发:不知道用哪个流程、有哪些工作流、怎么派活、组队做、一键跑完、多角色协作、agent 工作流怎么起、帮我选个流程、这活该谁干、从哪开始|English: which workflow, workflow router, route to agent pipeline, how to start, pick a skill|23 件 Agent 工作流技能的路由入口层:按意图把请求接到对应技能,固化派活前置三问、档位现场确认、消息纪律、工作区纪律、收尾沉淀五条硬协议;没点技能名先加载本技能选路再加载目标技能;对不上走 agent-general-route。
- wangyuanchuan2022 SkillOpensource Sanitize
触发:发布前扫描、脱敏检查、推送把关、publish-scan、开源前把关|English: publish scan, sanitize before publishing, secret scan|发布前脱敏机检与卫生巡检:六类扫描+三值裁定(single CRITICAL=FAIL),配套 publish-scan.mjs;含 repo-onboard 第 3 阶段硬门。
- wangyuanchuan2022 SkillDsh Troubleshooting
触发:DSH 排障、没反应、假死、权限被拒、EPERM、乱码、子会话不动、无输出|English: DSH troubleshooting, no output, permission denied, EPERM, mojibake|DSH 运行排障症状索引(Symptom→Causes→Diagnostics):假 idle 与 429、管道 EPERM、PS5 管道污染、沙箱权限坑等已实证症状快速定位
- wangyuanchuan2022 SkillAgent General Route
触发:没匹配到流水线、通用路由、走通用流程、没有对应技能、其他技能都不适用、直接干就行、不确定该用哪个 agent、兜底派活、随便处理一下|English: general route, fallback route, no matching pipeline, just do it, default dispatch, generic dispatch|路由表选不出技能时的兜底入口:按体量×并行×上下文×风险四轴裁「派席 or 主会话单干」,双通道分配人格(方法论走 agent_spawn 角色库;执行创作走 de_session 临时 persona)。
- wangyuanchuan2022 SkillAgent Review Squad
触发:评审改动、代码审查、code review、帮我审一下、合并前检查、安全审查、挑毛病|English: code review, review my changes, pre-merge check, security review|评审小队(编排 4 个 sub agent):code-reviewer 与 security-reviewer 对同一 diff 并行盲评(P0-P2+文件:行号),critic 最终质量闸门(并指出缺什么),可选 code-simplifier;只读评审+组织者合成,落盘双条件核销。
- wangyuanchuan2022 SkillVerification Loop
触发:收尾核验、完工检查、六相验证、验证报告、交付前验证、VERIFICATION REPORT|English: verification loop, six-phase verification, verify before done, verification report|验证环路(verifier 检查表):完工自报前六相 build/type/lint/test/security/diff 逐相过门,输出固定 VERIFICATION REPORT;Build 失败即 STOP;跳过项单列不计通过。
- wangyuanchuan2022 SkillSubagent Clarify
触发:子会话缺参数、澄清协议、子代理问主会话、任务书澄清模式、缺信息怎么办|English: subagent clarification, missing parameters, ask the organizer, clarify|子会话缺信息时不得瞎猜、不得编造、不得擅自终止——按结构化提问→de_broadcast(wake:true) 问主会话→主会话问用户→答案回传→超时降级缺省表链路;含防编造条款
- wangyuanchuan2022 SkillSkill Stocktake
触发:盘点技能、技能审查、技能质量、该退休哪些技能、技能库存、retire 技能|English: skill stocktake, audit skills, skill quality review, quick scan|技能质量盘点:Quick Scan 与 Full Stocktake 两模式,批评席按检查表给五态裁决(Keep/Improve/Update/Retire/Merge);results.json 缓存+断点续跑。
- wangyuanchuan2022 SkillSession Handoff
触发:会话交接、换会话继续、跨天收工、上下文要满了、写交接文档、接手别人的活|English: session handoff, continue in a new session, context running out, write a handoff, resume from a handoff, absorb a handoff|DSH 会话交接:把工作状态提炼成八节交接文档供接手会话全量吸收(八节格式见正文);含多份交接文档候选排序判据;不自建存储,只承担意图与判断层。
- wangyuanchuan2022 SkillAI Slop Cleaner
触发:清理代码、去 slop、deslop、代码太臃肿、过度工程清理、重复代码清理|English: clean up code, de-slop, remove duplication, over-engineering cleanup|AI 屎山清理:对能跑但臃肿/重复/过度抽象的代码回归安全清理——锁行为→计划先行→五类分类→单味道单批次遍+重跑验证→质量门→writer/reviewer 分离禁自批→范围锁定→证据报告收尾。
- wangyuanchuan2022 SkillAgent Autopilot
触发:自动实现整个项目、全自动做完、从想法到代码、帮我一条龙做出来、多阶段项目开发、端到端实现|English: autopilot, build the whole project, end-to-end implementation, idea to code, full pipeline|项目全流程自动实现(编排 23 个 sub agent):六阶段想法→验证过的代码(Phase 0 扩写..5 收尾);硬门:计划需用户批准、验证不过不进下一阶段;已有 requirement-interview spec 或 consensus-plan 计划可跳对应前置。
- wangyuanchuan2022 SkillError Handling
触发:错误处理、重试策略、熔断、错误类型设计、错误边界、用户可见报错|English: error handling, retries, circuit breaker, error types, error boundary|Robust error handling patterns for TypeScript, Python, and Go
- wangyuanchuan2022 SkillConsensus Plan
触发:规划方案、方案评审、出个计划、重大方案先评审、架构方案讨论、计划要评审|English: plan review, design review, consensus plan, architecture decision|共识规划:规划与执行严格分离——前置门拦截欠规格执行请求;Planner 出固定快照计划,Architect 与 Critic 串行独立评审(≤5 轮至 APPROVE);pending approval 期间不改源码不 commit 不 spawn。
- wangyuanchuan2022 SkillAgent Feature
触发:加个功能、实现一个功能、新增特性、给项目加 X、做个小需求、feature 流水线|English: add a feature, implement X, new feature pipeline, feature development|特性开发流水线(编排 6 个 sub agent):analyst 验收→planner 计划→executor 实现→test-engineer 跑绿→code-reviewer 评审→verifier 取证;适用于给现有项目加明确新特性(需求清晰、不需 autopilot)。
- wangyuanchuan2022 SkillTdd Workflow
触发:测试先行、TDD、红绿循环、先写测试、RED 门、覆盖率核验、写测试、证据报告|English: TDD, test-first, red-green, RED gate, coverage, evidence report|TDD 手册(test-engineer/verifier 引用):强制测试先行——RED 门不可让渡(只写未执行不算 RED),GREEN 与覆盖率核验后写证据报告(五列保证表)。
- wangyuanchuan2022 SkillAgent Bugfix
触发:修个 bug、报错了、行为不对、构建失败、测试挂了、修完验证、定位根因、别再复现|English: fix this bug, broken build, failing test, debug, root cause, regression lock|缺陷追猎流水线(编排 5 个 sub agent):explore 定位→debugger 根因与最小修复(疑难升 tracer)→test-engineer 回归锁→verifier 取证;适用于报错、行为不符、构建失败等缺陷。
- wangyuanchuan2022 SkillSkill Eval
触发:沉淀技能、写个技能、技能入库、要不要建技能、技能重复、Absorb 还是新建|English: save a skill, skill quality gate, dedupe against existing skills, absorb into existing skill, before skill_manage create|技能沉淀前质量闸:skill_manage create 前先 grep 重叠检查→四态裁决(Save/Improve/Absorb/Drop)→Save 后写后校验,任一失败不报成功;memory-consolidate 的姊妹技能。
- wangyuanchuan2022 SkillAgent Loop
触发:长任务循环、心跳巡检、循环卡死、熔断、DAG 编排、自主循环、拆任务并行跑、设计/写/查/体检 loop、loop 跑飞、loop 快照、退出码、清理遍、循环硬门、选型树、可判定目标、五个崩法、plan build judge|English: autonomous loop, long-running task, heartbeat, circuit breaker, DAG orchestration, loop status, loop guard, watchdog, stale loop detection, stop condition, design/write/review a loop, runaway loop, decidable goal, plan/build/judge, sequential pipeline, de-sloppify pass, bounded loop, merge queue|DSH 自主循环三层合一(机制+判断+模式):loop-start 三查与停止条件、loop-status 快照与退出码、guard 3 轮熔断;四条件门;顺序流水线、De-Sloppify、RFC 驱动 DAG。
- wangyuanchuan2022 SkillAdvisor
触发:advisor、顾问、每轮评审、每回合把关、把控全局走向、自动评审、帮我盯着、旁听评审、开个顾问、每轮有人把关|English: advisor, review each turn, per-turn review, automatic review, oversight, second opinion, keep an eye on it|advisor 顾问:每 turn 结束自动用会话手动指定的模型评审主会话(独立上下文只读),分级建议 info/nit/concern/blocker,nit 及以上默认 steer 打断;被动机制无工具面(不设 advisor* 工具),经 Advisor 面板启用;模型未指定即不评审(无兜底)。
- bohdan-basov BundleFind Leads
Leads for what you sell: decision makers at companies that show a buying signal and fit your brief, with why now. Use when the user asks for leads, clients, or who to reach out to.
- abdelrahman31 BundleOrchestrator Mode
Run the main session as a lightweight orchestrator that plans, delegates to subagents, and synthesizes, so the main context stays small and clean. Use this whenever a task is multi-step, touches many files, needs codebase exploration or research, involves running tests/builds with long output, or would otherwise pull large amounts of text into context. Also use it when the user says "orchestrate", "delegate", "use subagents", "fan out", "keep context clean", "don't flood the context", or asks for a large feature, refactor, migration, audit, or investigation across a codebase — even if they don't name this skill.
- boriemannetje SkillApex Notes Writing
Create, edit, import, delete, and connect Apex Notes (Apex Nodes) Markdown notes, or create named canvas lines, rectangles, circles, freehand strokes, and text through the workspace annotation sidecar. Preserve hierarchy, dates, legacy squares, and existing canvas content.
- vladosskins178 SkillAgent Ops Telegram
Talk to the user over Telegram while you run in a tmux session on a server. Use when prompts arrive prefixed with "[TG @user chat_id=... msg_id=...]", when you need to reply, report progress or send screenshots and files to the user's phone with `aot send`, or when the Telegram relay seems broken (messages not arriving, duplicates, 409 Conflict) and you need `aot doctor`, `aot status` or `aot logs`.
- aminfara BundleWriting Studio
Write or review professional, human-facing prose so the result is clear, well-structured, correctly shaped for its document type, and free of AI-sounding filler. Use this skill WHENEVER you are about to write something a person will read (a proposal, email, chat or Slack message, document or wiki page, blog post, announcement, status update, RFC or design doc, release notes, exec brief, incident report, peer performance feedback, RACI or roles-and-responsibilities matrix, program or project poster, PR/FAQ or working-backwards doc) OR when asked to review, improve, polish, critique, proofread, tighten, sharpen, or shorten any existing text. Prefer it over ad-hoc style fixes, especially for multi-paragraph or long-form writing. Do NOT use it for code, code comments, commit messages, machine-readable data (JSON/YAML/CSV), or trivial one-line replies.
- aminfara BundleSkill Linter
Lint a SKILL.md against the canonical structural template and report findings (missing sections, wrong ordering, non-canonical headings, infinite-loop risks, polyglot skills) for skill-creator to fix. Use this skill whenever the user asks to lint a skill, whenever a skill has just been created or edited with skill-creator, or whenever the user mentions skill structure, skill template, or skill consistency.
- aminfara BundleProgramming
Write, change, or fix code test-first in vertical slices, following one set of programming and testing principles (deep modules tested through their interface, functional core and imperative shell, explicit dependencies, modular monolith, mocks only at system boundaries). Use whenever you are about to write or modify production code or its tests, add a feature, or fix a bug, in any language. It also says when a change needs no tests, such as throwaway scripts and operational glue with no business logic.
- keanehu1988-sudo BundleXiaomi Auto Livestream Review
小米汽车(SU7/YU7)门店抖音付费直播间的直播大屏复盘:读巨懂车/巨量直播大屏的漏斗与趋势曲线,下载回放转写话术,逐分钟对齐找出留资话术与掉人话术,打分并给优化建议。判定基准从用户配置读取。复盘小米汽车(或其他汽车)直播场次时使用。
- marcelo-barella BundleGroketective
Grok Bot Detective. Deep read-only forensics on Grok Bot (internal codename sand) local install, Electron user-data, ~/.grokbot, app.asar, and sand-client-persistence slices. Also imports recovered Windows markdown transcript zips into live bots via /groketective import. Writes .cursor/plans/groketective-<theme>.plan.md. Use when the user invokes /groketective, /groketective import, or asks to reverse-engineer Grok Bot internals or restore recovered Grok Bot chats.
- seismoforg SkillCommit
Propose how to split the current changes into commits, commit them, then merge the feature branch into main and push, each step after the user's yes. Use when the user wants to commit, merge or push, or when the plan skill closes a plan.