skills

Explicit-only, provider-agnostic controller for bounded convergence of an existing technical plan, architecture proposal, migration plan, ADR draft, or implementation proposal. Use when the user explicitly asks for 方案收敛, plan convergence, 方案棘轮, or a controlled review-revision loop, or when an already user-authorized parent workflow explicitly delegates that convergence step with artifact, scope, provider bindings, budget, and modification authority. It decides whether to obtain review, revise locally, gather evidence, request a decision, explore alternatives, pivot, finalize, or stop. Do not use for first-draft planning, standalone review, generic brainstorming, code repair, implementation, or open-ended autonomous loops.

Use when the user needs to summarize, classify, deduplicate, and triage business feedback from a bounded Feishu/Lark group chat window using a specified domain knowledge pack. It turns chat messages into evidence-backed feedback cases, separates usage questions, misuse/config issues, suspected bugs, confirmed bugs, feature requests, product gaps, documentation gaps, resolved and unresolved items, and produces a digest plus memory-update candidates. Default to read-only. Do not create Feishu Project items, reply to chats, or update accepted memory unless explicitly approved.

Use when the user needs to read, search, summarize, analyze, or safely write Feishu/Lark IM messages, group chats, direct messages, threads, reactions, message resources, or bounded realtime IM events in a development workflow. Default to read-only evidence collection; require explicit confirmation for sending messages, replying, pinning, urgent notifications, member/chat changes, or cross-system writes. Do not use for Feishu Project/Meegle work-item operations, domain-backed business feedback triage, or Feishu document editing unless this skill is only gathering chat evidence for another Feishu workflow.

High-signal code change review workflow for PRs, patches, staged/index changes, working-tree diffs, commit ranges, and file-scoped review. Use when the user asks to review concrete code changes for functional/user-impact risks, data model correctness, state ownership, compatibility, or maintainability. Handles partial staged hunks and user-selected files/commits by selecting the exact review target before reading the diff. Do not use for implementation-only requests, generic code explanation, release publishing, issue triage, or plan-only Linus-style strict-pragmatic critique; for strict critique of a concrete diff, use this skill only to establish the target before dbx-linus-review.

Strict pragmatic, evidence-driven technical review for code changes, architecture plans, data models, and implementation proposals. Use when the user explicitly requests Linus-style review, harsh/strict review, pragmatic critique, over-engineering judgment, merge/readiness judgment, or asks whether a technical plan or code change is good enough. Shares judgment principles with dbx-diff-review but uses a stricter artifact-agnostic critique loop. Do not use for ordinary code explanation, implementation-only requests, generic encouragement, interpersonal judgment, or normal diff review unless strict/pragmatic critique is explicitly requested.

Use when the user asks to coordinate a development workflow across Feishu Project/Meegle work items, Feishu documents, and optional Feishu/Lark IM evidence, such as reading a ticket, updating a technical plan, syncing implementation progress, writing release notes, or commenting back with validation evidence. This is an orchestration skill; delegate project operations to dbx-feishu-project, document operations to dbx-feishu-doc, and chat evidence to dbx-feishu-im. Do not use for single-domain Feishu operations or generic planning without external Feishu read/write.

Read-only Alibaba Cloud Simple Log Service / SLS log query controller. Use when the user asks to query, generate, optimize, execute, summarize, or diagnose Alibaba Cloud SLS / Log Service / Logstore logs through Aliyun CLI, GetLogsV2, SDK, or Observability MCP. Produces a safe query plan, bounded command/API execution, structured findings, caveats, and next drill-down steps. Do not use for generic log files not stored in SLS, code implementation, destructive SLS management, credential collection, broad production data export, cloud cost optimization, or paid AI observability tools unless explicitly approved.

Use when the user needs to read, summarize, create, append to, or safely update Feishu/Lark Docx, Wiki, or document-like resources in a development workflow. Prefer official lark-cli document operations and block-aware updates. Do not use for Feishu Project/Meegle work items, generic writing without a Feishu document target, or editing external documents without clear permission.

Use when the user needs to read, query, create, update, comment on, attach files to, or transition Feishu Project / Lark Project / Meegle work items such as story, ticket, bug, requirement, version, iteration, task, defect, or release work. Default to read-only until a concrete write target, exact change, and user approval are clear. Do not use for Feishu documents, generic project-management advice, or code implementation without an external Feishu Project operation.

Use when the user has a fuzzy product/software idea, feature request, issue draft, stakeholder ask, or pre-development discussion and wants to clarify it into precise, testable requirements before product judgment, design judgment, technical planning, or implementation. Produces a bounded requirement contract with user/job/context, state change, scope, non-goals, functional requirements, acceptance criteria, edge cases, assumptions, open decisions, readiness state, and handoff. Do not use for direct implementation, code review, product-worth judgment, design critique, pure brainstorming, generic PRD writing, or formal DBX software plan-first phases unless explicitly invoked as a handoff source.

Source-grounded reading controller for explicit URLs, PDFs, papers, docs, GitHub pages/files, local files, pasted text, or small source sets. Use when the user asks to read, summarize, skim, extract Markdown, deep-read, compare, evaluate, translate-for-understanding, or turn source material into a reusable note. Handles fetch/grounding, thinking, and optional capture as one bounded workflow. Do not use for ordinary factual Q&A without a source, noisy mixed inbox routing, durable learning/practice plans, viewpoint article writing, product/design/code review, direct implementation, or silent external writes.

Chinese-first viewpoint writing for technical blogs, personal essays, opinion pieces, public drafts, Markdown articles, and occasional English versions. Use when the user wants to generate, develop, edit, or file prose that expresses a clear judgment with insight, accurate facts, and reader resonance. Do not use for commit/PR writing, code comments, product/design judgment, interpersonal message strategy, academic claims without sources, or deceptive/fabricated experience.

当用户明确想通过 AI 提升自己对某个概念、技术主题、资料集或技能的长期理解与可迁移能力时使用。适用于系统学习、讲透一个概念、建立 mental model、深度研究、把资料转成练习、复习测验、学习路线、学习记录候选和可持续学习 workspace。默认轻量、无预配置、少打扰。不要用于普通事实查询、单篇普通总结、泛泛推荐、直接代码实现或调试、产品/设计/架构决策、skill 创建、一次性润色，或用户只想要答案而非学习过程的请求。

Evidence-grounded, read-only repository or module architecture health review for AI-coding-era codebases. Use when the user asks to audit architecture health, prevent repo decay, assess long-term maintainability, find source-of-truth drift, identify verification gaps, or evaluate whether a codebase is safe and effective for AI-assisted change. Do not use for concrete PR/diff review, strict Linus-style critique, implementation planning, direct code changes, release/publish checks, or skill portfolio audits.

Manual trigger only. Use only when the user explicitly names `dbx-software-plan-first-finalize-plan`, `$dbx-software-plan-first-finalize-plan`, or asks to manually trigger this exact DBX Software Plan-First plan finalization phase. Do not auto-trigger for ordinary plan writing, finalize, seal, or implementation requests.

Manual trigger only. Use only when the user explicitly names `dbx-software-plan-first-ground-plan`, `$dbx-software-plan-first-ground-plan`, or asks to manually trigger this exact DBX Software Plan-First read-only grounding phase. Do not auto-trigger for ordinary repo reading, fact checking, plan writing, or implementation requests.

Manual trigger only. Use only when the user explicitly names `dbx-software-plan-first-implement-feature`, `$dbx-software-plan-first-implement-feature`, or asks to manually trigger this exact DBX Software Plan-First review-gated single-task implementation phase. Do not auto-trigger for ordinary implementation, tasks.md, next-task, or plan-first requests.

Manual trigger only. Use only when the user explicitly names `dbx-software-plan-first-plan-issue`, `$dbx-software-plan-first-plan-issue`, or asks to manually trigger this exact DBX Software Plan-First plan convergence phase. Do not auto-trigger for ordinary planning, Plan mode, repository exploration, or implementation requests.

Manual trigger only. Use only when the user explicitly names `dbx-software-plan-first-showhand`, `$dbx-software-plan-first-showhand`, or asks to manually trigger this exact DBX Software Plan-First safe automatic execution phase. Do not auto-trigger for ordinary automatic execution, do-it-all, showhand, or plan-first requests.

Evidence-grounded technical implementation planning before code changes. Use when the user asks to plan a software feature, refactor, migration, bug-fix strategy, infrastructure change, validation strategy, rollout, or codebase modification before implementation. Produces a scoped plan contract with goal, non-goals, evidence boundary, source of truth, invariants, implementation slices, validation model, risks, and handoff. Do not use for direct implementation, concrete diff review, strict critique of an existing plan, product/design judgment, generic brainstorming, commit/PR writing, or the stateful dbx-software-plan-first phase chain unless explicitly requested.

Use when the user asks to judge, audit, critique, review, shape, redesign, or produce a design plan for a UI, flow, PRD, screenshot, prototype, live product, component, design system, or code-backed interface. Grounds design judgment in target user, task path, information architecture, visual hierarchy, interaction states, visual language, design system consistency, accessibility, responsive behavior, and evidence. May read PRDs, code, screenshots, and live products. May inspect read-only surfaces. Must not edit files, write implementation code, apply patches, install packages, or commit changes.

Use when the user asks to judge whether a product, feature, PRD, information architecture, content, interaction, UI flow, live product, implementation, roadmap, or competitor position is right, good, usable, valuable, coherent, worth building, or product-ready. Grounds judgment in target user, job/context, evidence, alternatives, critical path, trust, and implementation alignment. Ask blocking questions or collect evidence before judging when background is missing. Do not use for pure implementation, generic UI inspiration, ordinary code review without product judgment, generic market summaries, or unsupported guesses.

Use to create, critique, improve, evaluate, or triage reusable agent skills from recurring scenarios. Enforces fail-closed gates, ASCT control-surface mapping, placement decisions, dominant failure-mode analysis, domain substance discovery, patch-first improvement hypotheses, runner-compatible evals, and artifact validation. Do not use for one-off writing, generic summarization, unsafe surveillance, or doing the user's direct task instead of designing a skill.

Manual-only skill. Use only when the user explicitly names `$dbx-skill-portfolio-auditor`, asks to manually trigger this skill, or launches it from the skill UI. Audits installed or repository agent skills to decide global vs repo/project vs explicit-only vs disable/uninstall. Do not implicitly invoke for generic skill audit, context-bloat, stale-skill, or cleanup requests; ask whether to run this skill instead. Do not use for creating a new task skill; use dbx-skill-architect for skill design.

Use when the user wants a reusable attention-routing workflow for mixed inbox items, saved content, tasks, ideas, signals, courses, tools, notes, or external-system metadata. Routes each item by stable attention states, grounded in visible evidence and a user-approved profile. Use high-risk prompts only to guard or safety-stop direct action requests. Do not use for ordinary summarization, one-off recommendations, product-specific automation by default, or silent external writes.

Manual-only bounded code ratchet for concrete code changes. Use when the user explicitly asks to run code ratchet, 棘轮自修, 棘轮实现, review-repair-revalidation, or L2 ratchet on staged changes, working-tree diffs, selected files, commit ranges, or pasted patches. This skill may modify code. It orchestrates existing review and repair skills, triages findings, blocks direction failures, applies only bounded local repairs, validates, and stops with a pass or escalation report. Do not use for read-only review, open-ended implementation, product planning, background automation, or Ralph-style PRD completion loops.

Codex 专用的 subagent 上下文策略辅助技能。Use when 用户要求使用 Codex subagent delegation、Codex worker/explorer delegation、并行 review、独立 reviewer、隔离上下文、不要带上下文、fork_context=false，或把 review/dbx-linus-review/waza-check/plan-eng-review 等技能交给 Codex subagent 执行时。默认新开 Codex subagent 不继承父线程历史，只传最小任务摘要；只有用户明确要求继承父线程时才使用 fork_context=true。不要因为用户讨论业务代码里的 worker、browser worker、架构 explorer 概念而触发。

Use only when the user wants to hand off the current AI agent session state to another AI agent or a future AI session, including context compaction, restart packets, resume notes, or "下个 session/agent 继续". For ambiguous "handoff/交接文档" requests, use only to ask one clarification question and stop. Do not use for human-facing workplace handoff documents, onboarding docs, project transition memos, status reports, meeting summaries, or general summaries.

当用户面对高影响、不确定、有真实行动后果的决策，需要在多个选项或 go/no-go 之间取舍时使用。适用于职业去留、技术架构、产品方向、团队组织、生活迁移、投资框架等。先做门禁，再决定 full_decision、quick_decision、clarification、direct_answer 或 safety_redirect。不要用于单纯概念解释、翻译润色、机械代码、一次性写作、低风险日常小事或没有行动后果的泛泛分析。

Generate or review concise English commit messages and PR descriptions for public or open-source repositories. Use when the user asks for commit text, PR body, release-style change summary, or validation wording based on a final diff, staged changes, working tree changes, or a specific commit. Do not use for internal M-xxx work PR contracts, generic code review, or explanations of what commits are.

用中文生成或评审工作场景的 commit message 与 PR 描述。适用于用户请求根据最终 diff、 staged changes、working tree changes 或指定 commit 生成 `M-xxx(type): title` commit， 以及中文合同式 PR 四块结构。只以最终变更集为事实边界，忽略聊天过程、失败尝试、 被否决方案和临时调试。不要用于公开开源英文 PR、普通代码 review、实现任务或泛泛解释 commit 概念。

Diagnose communication intent, disagreement layer, correction timing, boundary needs, and response strategy for relationship, work, and conflict conversations.

Use when the user wants to create, refine, start, or audit an OpenAI Codex goal contract for coding work, especially long-running or validation-heavy tasks. Produces copy-pasteable /goal commands, optional GOAL.md files, and only for large or strict tasks a goal package directory with plan, acceptance, validation, and status files. Do not implement the coding task unless explicitly asked.