Service Diffguard: read-only PR review lenses for correctness, error visibility, comment truthfulness, test coverage, type/contract boundaries, and simplification across every language in this repo (Go, Vue/JS/SCSS, Rego, protobuf, SQL, Helm/YAML, shell). Use for PR review, pre-commit review, unstaged-diff review, or a targeted review of one of those aspects.

Use when adding a cross-cutting concern (OTEL tracing, logging, metrics, caching, auth) to a business domain, creating files under business/domain/*/extensions/*, or adding the extension seam (ExtBusiness/Extension) to a *bus package.

.NET / C# 二进制逆向。当目标是 .NET assembly（PE 头含 CLR、.exe/.dll 托管程序）、C# 编译产物（含 NativeAOT）、红队 Sharp* 工具（Rubeus / SharpHound / SharpHound 等）、.NET 混淆程序（ConfuserEx / SmartAssembly / Babel / Eazfuscator）、.NET loader / info-stealer / 套壳 malware 时使用。优先用 dnSpyEx + de4dot，需要 AI 直接操作时联动 dnSpy MCP。不用于纯 native 二进制（走 reverse-engineering / ida-reverse）。

Fine-tune MiniCPM5-1B into a LoRA adapter and convert it to a GGUF adapter that loads directly into llama.cpp / llama-server and the MiniCPM Desk Pet app's custom-LoRA upload. Use when the user wants "GGUF LoRA", "convert LoRA to GGUF", "convert_lora_to_gguf", a custom persona/skin for the desktop pet, "桌宠自定义 LoRA", "上传 LoRA 到桌宠", or asks how to take a trained adapter and run it on a GGUF base.

Run MiniCPM5-2B or MiniCPM5-1B on-device with Google's LiteRT-LM runtime — the litert-lm CLI or its OpenAI-compatible server on a desktop, the Kotlin API or the AI Edge Gallery app on Android, the same .litertlm bundle on CPU or GPU. Use when the user says "LiteRT", "LiteRT-LM", "litertlm", ".litertlm", "Android", "Edge Gallery", "on-device app", or wants one bundle for phones and desktops.

Prepare for due diligence meetings — management presentations, expert network calls, customer references, and advisor sessions. Generates targeted question lists, benchmarks to reference, and red flags to probe. Use before any diligence meeting or call. Triggers on "prep for management meeting", "diligence call prep", "expert call questions", "customer reference questions", or "meeting prep for [company]".

Track multiple live deals with milestones, deadlines, action items, and status updates. Maintains a deal pipeline view and surfaces upcoming deadlines and overdue items. Use when managing a book of business, tracking process milestones, or preparing for weekly deal reviews. Triggers on "deal tracker", "deal status", "where are we on", "process update", "deal pipeline", or "weekly deal review".

Identify, measure, and exclude bot / crawler / AI-agent traffic in PostHog web and product analytics using the traffic classification surface (the isLikelyBot / getTrafficType HogQL functions and the $virt_* virtual properties). Use when the user asks to "exclude bots", "filter out crawlers", "remove bot traffic from my numbers", "how much of my traffic is bots / AI crawlers", "is GPTBot / ChatGPT / Claude hitting my site", "break down traffic by human vs bot", or wants clean human-only counts in an insight or dashboard. For the real-time Live tab bot tiles, use exploring-live-traffic instead.

Use when the user asks about revenue, payments, subscriptions, billing, CRM deals, support tickets, ad spend, production database tables, or other data PostHog does not collect natively — or wants to join or correlate PostHog product events with that external business data. Also use when a query fails because a table does not exist or returns no results for expected external data. The data warehouse can import from SaaS tools (Stripe, Hubspot, Zendesk, etc.), ad platforms, production databases (Postgres, MySQL, BigQuery, Snowflake), and other arbitrary data sources. Covers checking existing sources, identifying the right source type, and guiding the setup.

Guides exploration of $autocapture events captured by posthog-js to understand user interactions, find CSS selectors (especially data-attr attributes), evaluate selector uniqueness, query matching clicks ad-hoc, and create actions. Use when the user asks about autocapture data, wants to find what users are clicking, needs to build actions from click events, asks about elements_chain, wants to build a trend or funnel filtered by clicks or other autocapture interactions, asks which properties autocapture sends, or asks how to filter $autocapture events. Only applies to projects using posthog-js autocapture.

Use when adding or editing a GitHub Actions workflow that pushes a container image to a registry (ECR/ghcr/Docker Hub via build-push-action) or dispatches a production deploy (a `commit_state_update` repository_dispatch to PostHog/charts). Those run from a single canonical deploy repo, gated by the CD_DEPLOY_ENABLED variable. Does NOT apply to workflows that publish GitHub releases, npm, crates, or Homebrew — those stay on the public repo.

Produce and structure slow-query performance reports for PostHog's production ClickHouse (US and EU). Use when asked for a slow query report, query performance analysis over the last N days, per-team query cost, OOM or timeout investigation, cluster cost/memory regressions, or materialization candidates. Covers the modern `query_log_archive` source (typed `lc_*` columns, multi-day retention), how to categorize and attribute slow queries, root-cause patterns (unmaterialized JSONExtract, high-cardinality breakdowns, heavy joins), and the report structure. Runs queries via the `querying-production-databases-via-metabase` skill.

Guide for writing or reviewing PostHog kea logic files (`*Logic.ts` / `*Logic.tsx`). Use when creating a new logic, adding actions/reducers/selectors/listeners/loaders/forms/router bindings, choosing between reducer vs selector vs cache, deciding between listeners and `kea-subscriptions`, wiring React with `useValues`/`useActions`/`BindLogic`, or onboarding to kea conventions. Read keajs.org for upstream API; this skill captures PostHog-specific conventions and idioms.

Keep heavy imports off the django.setup() path that every process (web, celery, temporal, migrate, shell, CI) pays for. Use when touching AppConfig.ready(), wiring signal receivers, editing the lazy API router (posthog/api/rest_router.py or its __init__.py shim), deferring a heavy import, when the startup-import-budget guard fails, or when merging master into a long-lived branch that made the router lazy.

Generate beautiful code snippet images (PNG) for social media, tweets, LinkedIn, and documentation. Creates light-themed, macOS-style window chrome screenshots from code using HTML templates and Chrome headless. Use when asked to create code images, tweet images, snippet screenshots, code cards, or visual code examples for sharing. Supports single snippets, multi-image sets, and 2x2 grid composites.

Follow RSS feeds and curate them with LLM-scored relevance. Builds a searchable SQLite knowledge base, produces ranked daily digests, and learns your interests from feedback. Use when the user wants to check their feeds, search past articles, give feedback, star articles, or update their interest profile. Triggers on: "rss digest", "what's new in my feeds", "search my feeds for X", "thumbs up on #3", "star #1", "update rss profile", "/rss-curation", "rss search", "curate my feeds", "show starred articles".

Follow RSS feeds and curate them with LLM-scored relevance. Builds a searchable SQLite knowledge base, produces ranked daily digests, and learns your interests from feedback. Use when the user wants to check their feeds, search past articles, give feedback, star articles, or update their interest profile. Triggers on: "rss digest", "what's new in my feeds", "search my feeds for X", "thumbs up on #3", "star #1", "update rss profile", "/rss-curation", "rss search", "curate my feeds", "show starred articles".

SkillsBench task authoring — walk a contributor from idea to submission-ready task following CONTRIBUTING.md and the task-implementation rubric. Use when the user wants to create a new SkillsBench task, scaffold a task from an existing workflow (notebook, Excel workbook, document, dataset), convert a prompt or a benchmark item into a SkillsBench task, write skills for a task, or prepare a SkillsBench PR. Pairs with `task-review` (run that as a self-check before submitting).

Produce a polished, self-contained HTML "readout" document under ~/.readouts (with an auto-maintained index page), either by snapshotting the findings accumulated in the current conversation or — when invoked fresh, e.g. "/readout on how github webhook events are processed" — by sharpening scope with clarifying questions and researching the codebase before documenting. The work runs in a child agent so the main conversation's context stays clean. Use whenever the user invokes /readout, says "write this up", "turn this into a doc/page", "make a readout", or asks for a readable, shareable document capturing findings or explaining how something works.

Model Kotlin persistence code correctly for Spring Data JPA and Hibernate. Covers entity design, identity and equality, uniqueness constraints, relationships, fetch plans, and common ORM (Object-Relational Mapping) traps specific to Kotlin. Use when creating or reviewing JPA (Java Persistence API) entities, diagnosing N+1 or LazyInitializationException, placing indexes and uniqueness rules, or preventing Kotlin-specific bugs such as data class entities and broken equals/hashCode.