All authors
Claude Skills by YSAA1
github.com/YSAA134 skills2 installs42 views
- AutoresearchResearch-loop controller: the user assigns a goal or open question (the approach may be completely unknown) and this skill runs it independently as a repeated research -> hands-on -> adversarial review -> distill loop — investigate, run experiments, attack your own conclusions, distill; check completion criteria and, if not met, re-open by exclusion until one of five honest terminal states (answered/inconclusive/blocked/budget-exhausted/cancelled). Triggers: autoresearch, 自动研究, 研究循环, research...Votes: 0GitHub stars: 3
- BrainstormApproach brainstorming: grill a vague requirement into a settled approach, then a user-approved Spec. Takes a depth gear (low = few questions, fast out / medium = the daily default / high = research-grade deep grilling for bottlenecks and innovative design). Triggers: brainstorm, grill, discuss first, spec first (先讨论/grill/先落 Spec) — use when approach/goals/boundaries/trade-offs/success criteria/verification strategy are unsettled. Skip when a complete Spec already exists or for a small patch...Votes: 0GitHub stars: 3
- Capability RecommenderRecommends skills, hooks, MCP servers, subagents, or plugins for a confirmed tooling or workflow gap. Read-only recommendations only; when existing capabilities suffice, no tools are added, and nothing is auto-recommended for installation based on tech-stack keywords.Votes: 0GitHub stars: 3
- CleanupTidy the documents, code, generated artifacts, temporary files, and recovery state affected by this task; supports completion, blockage, or handoff. Ends with zero modifications when checks show no drift; does not expand into repo-wide cleanup.Votes: 0GitHub stars: 3
- DiagnoseInvestigate runtime, build, test, or performance failures whose root cause is unknown. Locate the cause via logs, reproduction, comparison, and falsifiable hypotheses; a known localized fix needs no full diagnostic process.Votes: 0GitHub stars: 3
- Find SkillsUse when the user explicitly looks for a reusable skill, or a confirmed capability gap is well suited to being filled by a skill. Ordinary how-to questions, code changes, or fact lookups do not automatically trigger a skill search.Votes: 0GitHub stars: 3
- HandoffCompact the current session into a handoff document so a brand-new session or another agent can take over losslessly. Use when the session is ending but the work is unfinished, when switching sessions/agents to continue, or when assigning a directed task to the next session. Triggers: handoff (交接, 会话交接, 换会话, 移交).Votes: 0GitHub stars: 3
- Harness BuilderBuild or repair the agent workbench in a target project: project entry, recovery surface initialization, verification gates, capability configuration. Triggers: onboarding this skill set into a new project for the first time, or the user asks to build/initialize/repair the workbench; routine tasks do not pass through (task-level setup is done automatically by plan); when the existing workbench is sufficient, return directly to the original task.Votes: 0GitHub stars: 3
- ImplementExecute well-scoped code, configuration, or documentation changes; behavior changes are test-driven by default (red first, then green). Verify according to risk and complete the authorized work; use diagnose for complex unknown failures, and clarify first when important design trade-offs are undecided.Votes: 0GitHub stars: 3
- PlanFor multi-stage tasks, tasks with complex dependencies, or tasks that need cross-session recovery, form an executable plan and, when the project has no recovery surface, establish a minimal recovery surface (work_index + state + AGENTS pointer). When the user only wants a plan, deliver it and stop; when execution is already authorized, continue after planning. Not mandatory for small patches.Votes: 0GitHub stars: 3
- Remove Deadcode PyOn-demand Python dead-code removal (repo-wide scope, not task close-out): Ruff fast lane + vulture tiered detection + reference/dynamic-access conviction + batched atomic deletion. Triggers: remove dead code, dead code python, unused code python (死代码清理, Python 清理死代码, 删除未用代码).Votes: 0GitHub stars: 3
- ResearchLightweight single-shot research: a background agent does the legwork and reads primary sources, filing the findings with per-item citations into a single markdown file. Fits one-off lookups (doc/API facts, topic scoping, research errands); for open questions that need multi-round hypothesis verification, use autoresearch. Triggers: research (调研, 查证, 帮我查资料, 一次性调研).Votes: 0GitHub stars: 3
- ReviewReview diffs, proposals, or completion claims, checking correctness, risk, and acceptance evidence. Supports limited-scope review of WIP; simple changes can be self-reviewed by implement, and verify is an alias of this skill.Votes: 0GitHub stars: 3
- ShipEnd-to-end delivery of an authorized, clearly scoped task: runs the three skills $implement, $review, $cleanup in sequence. Not used when scope or acceptance is unclear or the root cause is unknown; adds no new gates or repeated confirmations.Votes: 0GitHub stars: 3
- SweepOn-demand repo-wide reconciliation (not a task close-out): recovery-surface detection → inline reconciliation checklist → six-bucket classification → mechanical items auto-fixed, ambiguous items listed with evidence for human adjudication → four retirement steps → report. Triggers: sweep, take stock, big cleanup, project reconciliation (盘点, 大扫除, 项目对账).Votes: 0GitHub stars: 3
- TddTest-driven development discipline. Triggers: a behavior change calls for test-first, the user mentions red-green-refactor or integration tests (集成测试), or implement drives the test loop on pre-agreed seams.Votes: 0GitHub stars: 3
- Writing For AgentsEdits 'text written for agents to read': writing and revising skills themselves; auditing and revising durable instructions such as AGENTS.md/CLAUDE.md/Cursor rules; repairing and migrating the .harness recovery surface (work_index/state/lessons). Triggers: the user asks to write/revise a skill (写/改 skill), update or audit durable instructions (更新或审计持久指令), or repair/migrate a recovery surface (修复迁移恢复面). Division of labor — building the workbench in a target project goes to harness-builder; th...Votes: 0GitHub stars: 3
- Autoresearch研究循环总控：用户交办目标或开放问题（方案可以完全未知），独立跑「研究→干活→审核→蒸馏」来回循环——调研取证、动手实验、对抗审查、总结蒸馏，对照完成判据没完成就排除法续轮，直到五个诚实终态（answered/inconclusive/blocked/budget-exhausted/cancelled）。Triggers: autoresearch, 自动研究, 研究循环, research loop, hypothesis loop.Votes: 0GitHub stars: 3
- Brainstorm方案头脑风暴：把模糊需求聊成定型的方案，收敛成用户批准的 Spec。可带档位（低=少问快出 / 中=默认 / 高=研究性深挖，针对瓶颈与创新设计）。触发：方案/目标/边界/取舍/成功标准/验证策略未定，或用户说先讨论/grill/先落 Spec。已有完整 Spec 或小补丁时不用；批准后交给 plan。Votes: 0GitHub stars: 3
- Capability Recommender针对已确认的工具或工作流缺口推荐 skills、hooks、MCP、subagents 或插件。仅做只读推荐；已有能力足够时不添加工具，不按技术栈关键词自动推荐安装。Votes: 0GitHub stars: 3
- Cleanup整理本次任务影响的文档、代码、生成物、临时文件和恢复状态，支持完成、阻塞或交接。检查无漂移可零修改结束，不扩展为全仓清理。Votes: 0GitHub stars: 3
- Diagnose调查根因未知的运行、构建、测试或性能故障。以日志、复现、对照和可证伪假设定位原因；已知局部修复无需完整诊断流程。Votes: 0GitHub stars: 3
- Find Skills用户明确寻找可复用技能，或已确认的能力缺口适合通过技能补足时使用。普通如何做、改代码或事实查询不自动触发技能搜索。Votes: 0GitHub stars: 3
- Handoff把当前会话压缩成交接文档，让全新会话或另一个 agent 无损接手。会话要结束但工作未完、要换会话/换 agent 继续、要给下一个 session 定向任务时使用。Triggers: handoff, 交接, 会话交接, 换会话, 移交.Votes: 0GitHub stars: 3
- Harness Builder在目标项目搭或修 agent 工作台：项目入口、恢复面初始化、验证闸门、能力配置。触发：新项目第一次接入本套技能，或用户要搭/初始化/修工作台；日常任务不经过（任务级建档由 plan 自动做），已有工作台足够时直接回原任务。Votes: 0GitHub stars: 3
- Implement执行范围明确的代码、配置或文档修改，行为改动默认测试驱动（先红后绿）。按风险验证并完成已授权工作；复杂未知故障用 diagnose，重要设计取舍未定时先澄清。Votes: 0GitHub stars: 3
- Plan用于多阶段、依赖复杂或需要跨会话恢复的任务，形成可执行计划，并在项目无恢复面时建立最小恢复面（work_index + state + AGENTS 指针）。用户只要计划时交付后结束；已授权执行时计划后继续。小补丁不强制使用。Votes: 0GitHub stars: 3
- Remove Deadcode Py按需清除 Python 项目的死代码（全仓范围，非任务收尾）：Ruff 快通道 + vulture 分级检测 + 引用/动态访问定罪 + 分批原子删除。Triggers: remove dead code, dead code python, unused code python, 死代码清理, Python 清理死代码, 删除未用代码.Votes: 0GitHub stars: 3
- Research轻量单发调研：后台代理跑腿读一手来源，把结论连同逐条出处落成一份 markdown。适合一次性查证（文档/API 事实、主题摸底、查资料腿活）；开放问题需多轮假设验证时用 autoresearch。Triggers: research, 调研, 查证, 帮我查资料, 一次性调研.Votes: 0GitHub stars: 3
- Review审查 diff、方案或完成声明，检查正确性、风险和验收证据。支持 WIP 的有限范围审阅；简单改动可由 implement 自审，verify 为本技能别名。Votes: 0GitHub stars: 3
- Ship对已授权的范围明确任务做端到端交付：依次执行 $implement、$review、$cleanup三个skill。范围或验收不清、根因未知时不使用；不新增闸门或重复确认。Votes: 0GitHub stars: 3
- Sweep按需全仓对账（非任务收尾）：恢复面探测 → 内联对账清单 → 六档分类 → 机械项自动修、模糊项列证据人裁 → 退休四步 → 汇报。Triggers: sweep, 盘点, 大扫除, 项目对账.Votes: 0GitHub stars: 3
- Tdd测试驱动开发纪律。触发：行为改动要 test-first，用户提到 red-green-refactor 或集成测试，或 implement 在预先商定的 seams 上驱动测试循环。Votes: 0GitHub stars: 3
- Writing For Agents改「给 agent 读的文字」：skills 本身的写作与修订、AGENTS.md/CLAUDE.md/Cursor rules 等持久指令的审计修订、.harness 恢复面（work_index/state/lessons）的修复与迁移。触发：用户要求写/改 skill、更新或审计持久指令、修复迁移恢复面；分工——在目标项目搭工作台用 harness-builder，本 skill 只管文字与协议。Votes: 0GitHub stars: 3