.agents

automate long-running sampling of android anon 16KB large folio fallback stats via adb; run memstress workload and output raw/derived csv + summary.

Use when a Linux kernel task is primarily about top-down analysis, path classification, tracepoint work, trace capture, large-log search with rg, or git bisect. This router chooses the correct analysis workflow before detailed code or log analysis starts.

Use when a Linux kernel task is primarily about QEMU or guest VM execution, including starting or reusing a VM, guest SSH, QGA, reproducing issues inside the guest, or running xfstests or fsstress in the VM. This router chooses the correct VM-side workflow before detailed debugging starts.

Use for Linux kernel development workflows that may involve multiple related skills, especially when the user mentions QEMU or guest VMs, top-down analysis, tracepoints, large log search with rg, git bisect, long-running experiment capture, or handoff-style reports. Read this router first, then choose the relevant sub-router or workflow.

妙搭（Spark/Miaoda）应用开发与托管：应用创建、HTML静态站点发布、本地全栈开发、云端生成迭代、AI相关能力和飞书平台能力或者其他外部能力集成、日志/Trace/监控指标/PV/UV 查询、环境变量管理。当用户要开发/新建一个系统·工具·平台·应用，或要本地开发 / 云端开发 / 修改 / 部署 / 发布 / 上线 / 拿可分享链接，或用 HTML 做页面·网站·部署到妙搭，或提到妙搭/Spark/Miaoda（应用运行时域名形如 *.aiforce.cloud）、应用数据库、应用文件存储、开放 API Key、可见范围、线上日志、接口请求量、错误量、延迟、访问量、环境变量时使用。不负责普通云盘文件上传（lark-drive）、飞书文档编辑（lark-doc）、原生幻灯片创建（lark-slides）。

飞书云文档（Docx / Wiki 文档）：读取和编辑飞书文档内容。当用户给出文档 URL 或 token，或需要查看、创建、编辑文档、插入或下载文档图片附件时使用。文档中嵌入的电子表格、多维表格、画板，先用本 skill 提取 token 再切到对应 skill。当用户给出 doubao.com 的 /docx/ 或 /wiki/ URL/token 时，也应直接使用本 skill；路由依据是 URL 路径模式和 token，而不是域名。不负责文档评论管理，也不负责表格或 Base 的数据操作。当用户明确要操作飞书思维笔记时，也使用本 skill。

飞书云空间（云盘/云存储）：管理 Drive 文件和文件夹，包含上传/下载、创建文件夹、复制/移动/删除、查看元数据、评论/权限/订阅、标题、版本和本地文件导入。用户需要整理云盘目录、处理云空间资源 URL/token，或导入 Word/Markdown/Excel/CSV/PPTX/.base 为 docx/sheet/bitable/slides 时使用；doubao.com 云空间 URL/token 也按资源路径和 token 路由，不回退 WebFetch。不负责：文档内容编辑（走 lark-doc）、表格/Base 表内数据操作（走 lark-sheets/lark-base）、知识空间节点/成员管理（走 lark-wiki）、原生 Markdown 文件读写/patch/diff（走 lark-markdown）。

飞书邮箱：Use when user mentions 起草邮件、写邮件、草稿、发送/回复/转发邮件、查阅邮件、看邮件、搜索邮件、邮件文件夹、邮件标签、邮件联系人、监听新邮件、邮件收信规则等；use for mail/email intent only. Do not use for docs/sheets/calendar/auth setup/pure contact lookup/IM chat tasks.

飞书幻灯片：创建和编辑幻灯片。创建演示文稿、读取幻灯片内容、管理幻灯片页面（创建、删除、读取、局部替换）。当用户需要创建或编辑幻灯片、读取或修改单个页面时使用。当用户给出 doubao.com 的 /slides/ URL/token 时，也应直接使用本 skill，不要因为域名不是飞书而回退到 WebFetch；路由依据是 URL 路径模式和 token，而不是域名。不负责：云文档内容编辑（走 lark-doc）、云文档里的独立画板对象（走 lark-whiteboard，注意 slide 内嵌的流程图/架构图仍属本 skill）、上传或下载普通文件（走 lark-drive）。

飞书视频会议会中能力：用于让应用机器人真实加入或离开正在进行的会议，并读取当前身份可见的会中事件、发送会中文本消息或会中表情。适用于用户询问正在开的会议发生了什么、谁在发言、是否共享内容，或需要发现当前可读的进行中会议 ID。不负责已结束会议搜索、参会人快照、纪要、逐字稿或录制查询，这些使用 lark-vc 技能。

飞书知识库：管理知识空间、空间成员和文档节点。创建和查询知识空间、查看和管理空间成员、管理节点层级结构、在知识库中组织文档和快捷方式。当用户需要在知识库中查找或创建文档、浏览知识空间结构、查看或管理空间成员、移动或复制节点时使用。当用户给出 doubao.com 的 /wiki/ URL/token 时，也应直接使用本 skill，不要因为域名不是飞书而回退到 WebFetch；路由依据是 URL 路径模式和 token，而不是域名。不负责：上传文件到知识库节点下（走 lark-drive）、编辑文档/表格/Base 内容（走 lark-doc / lark-sheets / lark-base）。

Diagnose AOSP/Android platform build failures and run Cuttlefish ART/runtime A/B workflows, including Soong/Siso/nsjail triage, incremental ART/Bionic APEX rebuilds, custom kernel/DLKM image pairing, permissive CVD launch, manual ADB proxy fallback, and 16KB-alignment A/B image validation.

Use when working in the Pixel common kernel tree on Android mTHP/order-0/deferred-split experiments, especially to distinguish active reason counters from archived smoke counters or to restore the archived UFFD order-2 mfill experiment.

use when the user wants to encrypt Codex or opencode API credentials at rest, launch either CLI through an encrypted key file, migrate away from plaintext auth.json, rotate keys without changing provider wiring, or keep the usual CLI UX while loading secrets from environment variables at launch

orchestrate safe and reproducible f2fs or qemu work inside a learn_os style kernel workspace. use when chatgpt needs to source .vars.sh, start qemu without blocking, validate kernel builds, run guest commands, run guest test scripts, verify f2fs/mounts, and collect logs with evidence. default to ssh when available, but if ssh is unavailable/blocked or the user explicitly requests qemu guest agent execution, use `scripts/qga_exec.py` as the primary guest command runner.

Use when syncing modified kernel code files from a full kernel tree to a lightweight shadow repository, including batch initialization, incremental sync, and auto-daemon workflows.

Design task-specific harnesses for long-running experiments, benchmarks, boot loops, stress tests, samplers, or agent-run automation. Use when Codex is about to create or run a long-run script, synthetic workload, Cuttlefish/QEMU/device experiment, A/B benchmark, sampling loop, or any workflow where early bad signals should stop the run instead of wasting time. The skill teaches Codex to derive minimal control parameters, reject unjustified sleeps/intervals/touch strides, add task-local self-check hooks, define early sanity gates, and provide a concrete implementation example without hardcoding one domain.

Install, repair, and smoke-test xfstests inside a QEMU Ubuntu guest when SSH is unavailable and all operations must run through qga_exec.py. Use this skill whenever the user mentions xfstests installation failures, missing /usr/local/bin/check, QGA-only guest control, loopback-based test setup, or asks for reproducible xfstests automation and troubleshooting playbooks.

generate correct, runnable adb commands and host-side bash scripts for android devices (pixel and others), including monkey automation, package discovery, crash artifact collection, and Android ART odex/vdex triage. use when a user asks for adb/adb shell commands, dexopt/oat/vdex/oatdump/vdexdump help, ClassNotFoundException or NoClassDefFoundError triage, package names, pulling crash artifacts, or avoiding common root/quoting/host-vs-device pitfalls.

Use when adding or restructuring Linux kernel debug logs, especially in Pixel/common where existing structured klog should be the default and pr_debug should only fill a specific missing edge.

Debug and repair Kiro CLI agent configurations when tools are printed as literal XML/text, tool calls never execute, approvals still appear after configuring trust, or reads/writes outside the workspace fail. Use when working with ~/.kiro/agents/*.json, ~/.kiro/settings/cli.json, Kiro CLI sessions under ~/.kiro/sessions/cli, or commands such as kiro-cli agent validate/list/set-default and kiro-cli chat --no-interactive.

对外交接文档/数据报告风格规范。强制要求从接收者视角写作，确保最小可复现命令、核心参数直观解释、指标先解释再展示，并禁止写入只有作者/AI知道的内部信息。

Permission-aware install and test workflow for Codex. Use this whenever the user asks to install, update, or test open-source skills, MCP servers, CLIs, browser tools, or package-managed runtimes and the task may trigger sandbox approvals, network access, writes outside the workspace, or repeated "allow this command" prompts. Also use when the user mentions permissions, white lists, prefix rules, sandbox interception, "why did this read-only command get blocked", or asks to reduce repetitive approvals.

Write office-style experiment reports and status-update modules in Markdown for different leadership audiences. Use this whenever the user asks for 技术汇报、阶段汇报、领导汇报、办公口吻文档、结论先行总结、技术结果说明、问题与下一步计划，especially when the same data needs different wording for technical leaders versus department leaders, or when the user wants one version for自己阅读 and one version for上报. Always produce a concise conclusion-first module, prefer charts before tables, and consult the matching file under roles/ before drafting.

Use when analyzing Android kernel mmap_lock or per-VMA lock contention, especially to validate rip-out patches, tracepoint capture workflows, and kernel or user stack collection on rooted Pixel devices.

Use when the user wants a persistent local MCP capability for secure `opencode` task execution, especially when extracting ad-hoc wrappers into a reusable bottom-layer service, running cheap external subagent work, or keeping encrypted-at-rest API keys while exposing `opencode` through stdio MCP.

Use when a thesis or paper is authored in Word and the agent needs to insert editable, thesis-style pseudocode blocks into an existing .docx without converting them to images

Archived Pixel common mTHP/order-0 experiment code from June 2026, including debugfs counters, tracepoints, swap birth records, VMA suitability attribution, UFFD/madvise/deferred-split instrumentation, and restore instructions. Use when asked to restore or inspect the archived high-overhead mTHP experiment patches.

Use when debugging a complex bug and you need a maintained hypothesis matrix whose statuses are updated only with executable code paths and exact file:line evidence, especially for kernel, filesystem, writeback, mmap, race, or corruption investigations where vague speculation is harmful.

Orchestrate iterative Linux kernel debugging in learn_os style workspaces by chaining QEMU bring-up, targeted instrumentation, build, test execution, and evidence-driven next-iteration decisions. Use this when tasks are kernel debugging loops (for example writeback/f2fs/fscrypt crashes), and you need a repeatable cycle that depends on f2fs-qemu-agent-pipeline plus kernel-log-instrumentor.

help with android kernel profiling on rooted devices, especially pixel phones and aosp or android common kernels. use when asked about simpleperf, perfetto, linux perf on android, kernel symbols, call stacks, vmlinux, ko modules, f2fs, aosp kernel build outputs, stripped binaries, or how to feed kernel debug artifacts into report_html.py. trigger for setup, measurement, comparison, troubleshooting, and command generation for user or kernel time split, flamegraphs, off-cpu traces, source annotation, and disassembly.

Use when adding or fixing Linux kernel ftrace tracepoints, especially TRACE_EVENT definitions, CREATE_TRACE_POINTS placement, caller includes, and common link or compile failures around trace event wiring.

Use when sending messages via lark-cli, configuring git sendemail, or any task that needs the user's Feishu bot appId or git SMTP identity. Provides the cached identity map so agents don't need to re-search.

Top-down recursive kernel diagnosis using percpu arrays for coarse-grained counting then tracepoints for fine-grained stack capture. Use when you need to classify and quantify a multi-path kernel code flow (e.g., deferred split reasons, page fault types, reclaim types) and want minimal overhead for routine counts while retaining the ability to drill into hot paths with full stack traces.

termux/android device onboarding and operation over adb forward plus ssh/scp for fio testing. use when the user wants chatgpt or an automation agent to add a new termux device, maintain a persistent device-to-port registry, allocate the next local adb forward port, generate ssh config entries, run remote commands, handle required human handoffs for adb/storage/root authorization, combine ssh with su safely, and pull fio results from android termux back to the host.

Use when the user requests diagrams, flowcharts, architecture diagrams, ER diagrams, UML / sequence / class diagrams, network topology, ML/DL model figures (Transformer/CNN/LSTM), mind maps, or any visualization. Also use proactively when explaining systems with 3+ components, complex data flows, or relationships that benefit from visual representation. Best suited when the diagram needs custom styling, rich shape vocabulary, swimlanes, or exportable images (PNG/SVG/PDF/JPG). Generates .drawio XML and exports locally via the native draw.io desktop CLI.

Deploy, evaluate, and manage Foundry agents end-to-end: Docker build, ACR push, hosted/prompt agent create, container start, batch eval, continuous eval, prompt optimizer workflows, agent.yaml, dataset curation from traces. USE FOR: deploy agent to Foundry, hosted agent, create agent, invoke agent, evaluate agent, run batch eval, continuous eval, continuous monitoring, continuous eval status, optimize prompt, improve prompt, prompt optimizer, optimize agent instructions, improve agent instructions, optimize system prompt, deploy model, Foundry project, RBAC, role assignment, permissions, quota, capacity, region, troubleshoot agent, deployment failure, create dataset from traces, dataset versioning, eval trending, create AI Services, Cognitive Services, create Foundry resource, provision resource, knowledge index, agent monitoring, customize deployment, onboard, availability. DO NOT USE FOR: Azure Functions, App Service, general Azure deploy (use azure-deploy), general Azure prep (use azure-prepare).

Use when needing to fetch Linux kernel patches or mailing list threads from lore.kernel.org, especially when encountering Anubis anti-bot protection, or when converting email patches to git-am ready format. Trigger for any task involving kernel patch retrieval, lore.kernel.org access, public-inbox mbox downloads, or b4 tool usage.

Use the rw/mmap test harness in /home/nzzhao/learn_os/myscripts/shared_with_qemu/test to run, extend, or refactor buffered I/O and mmap test cases around rw_test.py, rw_matrix.sh, rw_matrix_inline.sh, f2fs_gc_long_rw.py, mkwrite_test.py, and mmap_wp_fault_test.py. Trigger this whenever the user asks to add or modify read/write test cases, matrix cases, long-running GC/writeback stress, inlinecrypt or fscrypt coverage, hole/existing/truncate cases, read-then-write coverage, mmap fault cases, or wants agents to use this framework rather than inventing one-off scripts.

中文学术毕业论文（本科/硕士/博士）全流程写作指导与质量检查 Skill。 用于辅助毕业论文的撰写、修改、审查与定稿，覆盖写作风格、文献检索、 论文结构规范、排版语法、实验规范、盲审合规性、图表一致性等维度。 使用场景： (1) 撰写或修改论文章节内容时，确保符合学术写作规范 (2) 润色文本，去除 AI 写作痕迹，使行文自然如人类研究者亲笔 (3) 检索、筛选和管理学术参考文献 (4) 检查论文结构、逻辑一致性、盲审匿名性、图表风格一致性 (5) 定稿前的全面复盘检查