Explore

Detect potential code-level performance smells in Java - streams, collections, boxing, regex, object creation. Provides awareness, not absolutes - always measure before optimizing. For JPA/database performance, use jpa-patterns instead.

Audit Maven dependencies for outdated versions, security vulnerabilities, and conflicts. Use when user says "check dependencies", "audit dependencies", "outdated deps", or before releases.

Java logging best practices with SLF4J, structured logging (JSON), and MDC for request tracing. Includes AI-friendly log formats for Claude Code debugging. Use when user asks about logging, debugging application flow, or analyzing logs.

Guide for upgrading Java projects between major versions (8→11→17→21→25). Use when user says "upgrade Java", "migrate to Java 25", "update Java version", or when modernizing legacy projects.

Systematic code review for Java with null safety, exception handling, concurrency, and performance checks. Use when user says "review code", "check this PR", "code review", or before merging changes.

Triage and categorize GitHub issues with priority labels. Use when user says "triage issues", "check issues", "review open issues", or during regular maintenance of GitHub issue backlog.

Review Java concurrency code for thread safety, race conditions, deadlocks, and modern patterns (Virtual Threads, CompletableFuture, @Async). Use when user asks "check thread safety", "concurrency review", "async code review", or when reviewing multi-threaded code.

Clean Code principles (DRY, KISS, YAGNI), naming, function design and readability. Use when code is hard to read, with long methods, unclear names, duplication or deep nesting. For how responsibilities are split across classes and which way dependencies point, use solid-principles instead.

Deep need-clarification skill. Use only when the user explicitly invokes $task-clarifier. Once activated, keep asking until all three goals are met: the user fully understands their own needs, the AI fully understands the user's needs, and the user confirms the AI's understanding is correct. Do not intervene in task execution unless explicitly invoked.

Cooperatively pause unfinished work in the current session, leave a precise continuation checkpoint, and resume from it when the user later says continue. Use only when the user explicitly requests this pause workflow. Do not use for crashes, app shutdown, new-session handoffs, or ordinary task summaries.

Delegate a general-purpose task to Google's Antigravity CLI (agy staffer, fast Gemini) with a minimal, unopinionated prompt. Use when the user says /agy:staffer, "have agy do/handle X", "have agy generate an image", or the task fits none of the specialist personas (researcher / reviewer / implementer / ask) — the template adds no role, rules, or output format, so the task text alone shapes the output. Also the route to agy-native tools no specialist covers, notably image generation (generate_image).

Orchestrate an ongoing task with AGY while the current agent owns key decisions, review, and delivery. Use when the user invokes /agy:lead or asks you to coordinate a task using AGY.

Ask Google's Antigravity CLI (agy staffer, fast Gemini) a cheap one-shot question - the fast zero-tool mode and the post-install smoke test. Use when the user says /agy:ask, "ask agy", "quick second opinion from agy", or right after installing to verify the plugin works.

Delegate a general-purpose task to Google's Antigravity CLI (agy staffer, fast Gemini) with a minimal, unopinionated prompt. Use when the user says /agy-staffer, "have agy do/handle X", "have agy generate an image", or the task fits none of the specialist personas (researcher / reviewer / implementer / ask) — the template adds no role, rules, or output format, so the task text alone shapes the output. Also the route to agy-native tools no specialist covers, notably image generation (generate_image).

Get a second-opinion review from Google's Antigravity CLI (agy staffer, fast Gemini) - of code (a diff, PR, working tree) or of a decision, plan, or design. Use when the user says /agy-reviewer, "have agy review this", "second opinion on my diff/PR/plan", or after finishing work and wanting an independent verifier that does not share the host model's blind spots.

Delegate a deep research or survey task to Google's Antigravity CLI (agy staffer, fast Gemini). Use when the user says /agy-researcher, "ask agy to research", "have the agy staffer survey X", or wants a second, independent deep-dive on a topic or codebase without spending the host model's quota.

Orchestrate an ongoing task with AGY while the current agent owns key decisions, review, and delivery. Use when the user invokes /agy-lead or asks you to coordinate a task using AGY.

Manage agy staffer background jobs - collect results, check status, cancel, follow-up conversation, and setup. Use when an agy job needs collecting, when the user asks "is the agy job done", "show agy's result", "cancel the agy job", "continue the agy conversation", or "set up agy". This is the orchestrator's skill; the persona skills (staffer/researcher/reviewer/implementer) point here.

Delegate a coding task to Google's Antigravity CLI (agy staffer, fast Gemini), which edits the working tree directly and can perform explicitly requested Git delivery. Use when the user says /agy-implementer, "have agy fix/build X", or wants to hand a well-scoped coding task to the agy staffer instead of doing it in the host model.

Ask Google's Antigravity CLI (agy staffer, fast Gemini) a cheap one-shot question - the fast zero-tool mode and the post-install smoke test. Use when the user says /agy-ask, "ask agy", "quick second opinion from agy", or right after installing to verify the plugin works.

Design, change, or review an HTTP/JSON API surface — endpoints, request/response shapes, authentication and authorization, pagination, idempotency, rate limits, versioning, and deprecations. Use when adding or modifying an HTTP endpoint, reviewing an OpenAPI spec or HTTP route diff, or deciding whether an HTTP API change breaks consumers. Do not use as a protocol-compatibility checklist for gRPC/protobuf, GraphQL, WebSockets, or other non-HTTP/JSON interfaces.

Evidence-first development — surround the implementation with an executable spec and a gauntlet of constraints (tests, types, coverage, mutation) so line-by-line review becomes optional. Use when the user explicitly asks for high-assurance or evidence-first work ("reliable", "TDD", "prove it works", "I won't read the code"), or when the change touches high-stakes domains (money, auth, data loss, concurrency, public API). For routine changes where the user just wants normal tests, write good tests directly instead of invoking this loop.

Configure and use Duel Agents model routing in Claude Code. Requires a Duel API key from duelagents.com. Routes all LLM traffic through duelagents.com/v1.

Use when the user is unsure which Higgsfield workspace fits their task, needs to decide between Cinema Studio / Lipsync Studio / Draw-to-Video / Viral Hub presets / Click to Ad / Higgsfield Audio (Sora 2 Trends is built on the retired Sora 2), or is asking 'what should I use for X'. This sub-skill routes by production problem BEFORE model selection.

Use when the user asks about Vibe Motion, motion graphics, kinetic typography, animated text, brand animations, logo animations, infographic animation, presentation animation, data-driven video, or prompt-to-motion-graphic workflows. Vibe Motion is fundamentally different from video generation — it creates deterministic code-based motion graphics, not AI-predicted pixel sequences. Use this when the text/logo must stay perfectly crisp and editable (code). For a cinematic/kinetic AI-generated video ad built from a storyboard (pixel render, native audio), use higgsfield-motion-design instead.

Use when a Higgsfield generation fails, produces poor quality, looks wrong, doesn't match the prompt, or the user needs to fix or improve an output.

Use when the user asks about visual styles, aesthetics, color grades, film looks, or how to set the tone and atmosphere of a Higgsfield generation.

Use when the user mentions the Higgsfield CLI (binaries `higgsfield` / `higgs` / `hf`, `higgsfield auth login`, `higgsfield generate create`, `higgsfield workflow`, the `@higgsfield/cli` npm package), the Higgsfield MCP custom connector (`mcp.higgsfield.ai/mcp`) or its bundled workflows (`get_workflow_instructions` — character-sheet, thumbnail-generation, ugc-* videos, faceless-video, narrator, subtitles, product-photoshoot, brand-asset-creation, ad-multiplier, video-editing / Higgsedit, website-builder-flow), Higgsfield's bundled skills repo (`higgsfield-ai/skills` — `higgsfield-generate`, `higgsfield-soul-id`, `higgsfield-product-photoshoot`, `higgsfield-brandkit`, `higgsfield-marketplace-cards`, `higgsfield-websites`, `higgsfield-video-explainer`, `higgsfield-youtube-thumbnail`, invoked as `/higgsfield:generate` etc.), or asks how this skill coexists with those tools (`do I need both`, `how does this work with the CLI/MCP/skills`, `should Higgsfield's workflow do this or you`).

Creates and manages reusable character profiles (Soul IDs) for consistent facial and stylistic identity across multiple image and video generations. Provides identity-vs-motion prompt separation, character sheet creation workflows, micro-expression direction, and Soul Cast AI actor configuration. Use when the user wants to maintain character consistency across multiple generations, asks about Soul ID, creating reusable characters, or generating consistent people across different scenes and shots.

Turns a brief, script, scene breakdown, treatment, or story idea into ONE connected director's shotlist for Seedance 2.0 — a single editable HTML artifact with a global Style Prefix, an @-asset glossary, and named per-scene prompts (1a, 1b, 2a…) each in Style → Characters → Scene → CUT 1..N form. Use whenever the user says 'make a shotlist', 'break this script into prompts', 'generate Seedance prompts for this ad/film', 'turn this brief into a shot list', 'director's shotlist', or wants many connected scene prompts rather than one. Also use to revise an existing shotlist (edit-once-propagates: 'change the style prefix everywhere', 'rewrite scene 4', 'split prompt 6'). Each prompt targets 15s; longer scenes split across 1a/1b/1c under one scene number.

Writes, improves, or rewrites Seedance 2.0 prompts that TRANSFORM footage the user already has (video-to-video), rather than building a scene from scratch. Use whenever a real clip is the starting point and they want to: add a VFX element (set a head or hair on fire, transform a hand, make a limb invisible), swap the environment around a preserved subject (desert, clouds, lava, a neon city), drop a giant photoreal creature behind or onto a subject/landmark, relight or regrade so subject and added elements read as one shot, sync a crash-zoom or push-in to a spoken line or timecode, or generate a matching transformed start frame to animate from. Also use when they paste such a prompt and ask to change its lighting, timing, creature, or runtime, or say 'make a Seedance prompt for this video' with a clip attached. This is the video-to-video specialization; for a brand-new scene from image references with no source clip to preserve, higgsfield-seedance applies instead.

Seedance 2.5 prompt director — the omni-reference dialect. Routes the four generation modes (t2v / omni_reference / video_edit / video_extension), writes explicit @Image/@Video/@Audio reference roles with exclusions, stages 30-second videos into end-state beats, and covers video editing, forward/backward extension, first-last-frame and multi-keyframe control, storyboard grids, blockout rendering, and seamless transitions. Use whenever the user asks for a Seedance 2.5 prompt, mentions Seedance 2.5 / Dreamina / Jimeng, wants a clip longer than 15s on Seedance, wants to EDIT or EXTEND an existing video rather than generate a new one, or supplies more than a handful of image/video/audio references. For Seedance 2.0 (4K, `mode=fast`, or a genre hint) use higgsfield-seedance instead.

Rewrites scene descriptions using professional cinematography language, structures prompts with a six-slot formula (camera + subject + action + setting + style + lighting), and diagnoses content filter rejections via a preflight linter. Use whenever the user asks for a Seedance 2.0 / Seedance Pro prompt, describes a scene for Seedance generation, mentions Seedance, reports a Seedance generation failure or flagged prompt, or is burning credits on Seedance regenerations.

Tests whether a scene is structurally worth generating before any shot is prompted — a five-element engine of Goal, Obstacle, Tactic, Reversal, and Value Shift. Use when the user has a scene, sequence, beat outline, or script and wants it audited or strengthened; when a sequence generates cleanly but lands flat and nobody can say why; when shots look good individually but the run of them does not build; or when the user asks 'is this scene working', 'what's weak here', or 'why doesn't this land'. Upstream of prompting — it decides WHICH shots deserve the credits, not how to write them. Pairs with higgsfield-character-design (who the characters are) and higgsfield-shotlist-director (turning the settled scene into shots).

Use when the user already has a FINISHED video and wants Higgsfield to repurpose, restyle, clip or analyze it: Shorts Studio (restyle one 4–120s source video into a set of short-form clips from a style preset), Clipify (turn one YouTube video into subtitled clips), Virality Predictor (predict virality, hook strength, attention, retention risk with a dashboard), or Video Analysis (scene-by-scene breakdown of an uploaded video or a YouTube link). Also use when the user asks which of these cost credits and which are free, 'turn my video into shorts', 'will this go viral', 'score my ad', 'break this video down scene by scene'. Names: shorts_studio_create, shorts_studio_create_preset, shorts_studio_list_presets, shorts_studio_status, clipify, virality_predictor, brain_activity, video_analysis_create, video_analysis_status.

Use when the user wants a genre-specific scene template, starting point for a type of video (action, horror, romance, product ad, documentary, sci-fi, dance, etc.), or asks for prompt examples for a specific style of content.

Use this skill AUTOMATICALLY before writing any Higgsfield prompt. Query the memory databases for relevant past failures and pre-apply known fixes before the user even hits generate. Triggers include: any request to write a Higgsfield prompt, any use of the higgsfield-prompt skill, any mention of generating a video or image on Higgsfield, any MCSLA prompt construction. This skill should run SILENTLY in the background — don't announce it, just apply what's known. If the databases are empty, skip silently and proceed with normal prompt generation.

Use when building, writing, refining, or structuring a Higgsfield AI prompt. Covers the MCSLA formula, prompt structure, narrative vs. timestamped formats, and how to write for both text-to-video and image-to-video workflows.

Use when the user wants to create a complete multi-shot video, asks how to chain Higgsfield tools together, wants to build a short film or branded content series, asks "what's the full workflow", needs to connect Popcorn → image → video → Recast → audio → assembly, or wants to understand how the platform works as a production system rather than isolated tools.

End-to-end motion-design / animated-ad creation flow on Higgsfield via the MCP connector. Use when the user wants to create motion design, animate a logo, make a video from an image, build an animated ad or brand promo, turn a product into motion, or says 'make a motion', 'motion design', 'animate this', 'make a video from my logo', 'animated brand', 'motion graphics', 'brand motion', 'kinetic graphics', 'promo video', or 'ad video'. Drives a storyboard-first pipeline: brief → storyboard sheet (GPT Image 2) → video (Seedance 2.0) — an AI-generated pixel video clip. Distinct from higgsfield-motion (the named camera/motion preset library) and from higgsfield-vibe-motion (deterministic Remotion code with crisp text); use vibe-motion instead when the text/logo must stay perfectly crisp, editable, and deployable as code.

Use when the user wants to apply a named Higgsfield motion preset, asks about VFX presets, transformation effects, elemental effects, or transition presets. Contains the full named preset library with descriptions and prompt usage.

Use when the user asks about Moodboard, building a moodboard from reference images, curated moodboard presets, Soul Hex color transfer, applying a visual style direction to generations, or named moodboard styles like Y2K studio, Warm ambient, Theatrical light, Swag era, Flash editorial, Street photography, Asian nostalgia, Retro BW, Surreal solarization, etc.

Use when the user asks which model to use, wants to compare models, or needs guidance on selecting between Kling, Wan (incl. Wan 3.0), Seedance (incl. 2.5), Veo 3, Minimax Hailuo / MiniMax H3, FLUX 3 Video, Gemini Omni Flash, Soul, Nano Banana, GPT Image 2.5, or other Higgsfield engines — including which lane edits existing footage (video_edit, Kling 3.0 Omni Edit, FLUX 3 Video Edit, Genjutsu, Ad Multiplier) and which models make one clip longer than 15s.

Use when the user asks about Mixed Media, wants to apply artistic preset styles to an image (Noir, Sketch, Paper, Canvas, Particles, Neon, etc.), combine multiple artistic treatments, or create stylized non-photorealistic outputs.

Use when the user mentions Marketing Studio, DTC Ads, ad video, UGC video, the marketing_studio_video MCP model, or wants to generate one of the 9 Marketing Studio ad presets (UGC, Tutorial, Unboxing, Hyper Motion, Product Review, TV Spot, Wild Card, UGC Virtual Try On, Pro Virtual Try On). Also triggers on hook+setting picklist questions, preset-avatar / custom-avatar / text-generated-avatar handling, 4–15s ad video questions, or any reference to Higgsfield's ad-video product surface. Cross-surface workflow handoffs (GPT Image 2.0 / Soul Cinema / Nano Banana Pro / ms_image image gen → Marketing Studio video) covered in the companion cross-surface-workflow.md. Also use for Ad Multiplier (`ad_multiplier` — 'multiply my ad', many independently edited versions of one supplied 4–30s ad) and for Genjutsu one-off edits of a finished ad (swap one object / product / garment → `hf_mult_replace_object`; transfer motion → `hf_mult_motion_control`).

Use when the user wants to generate a cinematic still image on Higgsfield, asks about shot framing, camera angle, or composition for image prompts, needs a specific shot type (close-up, wide shot, cowboy shot, etc.), or is working with [img] reference-based image generation. Covers distance & size shots, camera angles, and camera movement keywords optimized for AI image models (Soul 2.0, Nano Banana, Seedream, GPT Image, Flux, etc.).

Use when the user mentions GPT Image 2.0 or GPT Image 2.5, gpt-image-2, gpt-image-2.5, gpt_image_2_5, GPT-Image-2 prompts, the Flare / Sunburst variants, a transparent-background image, or wants to generate an image with GPT Image 2.0 or 2.5. Covers when to prefer GPT Image 2.5 over 2.0 (transparent background, xhigh/max quality, extra aspect ratios — from platform and OpenAI facts), the CLI-vs-MCP disagreement on gpt_image_2, the three-format prompt taxonomy (Format A structured JSON for UI mockups and layout-dense images; Format B dense cinematic prose for single-subject scenes; Format C auto-derive meta-prompt for theme-only concepts), per-format craft patterns, output conventions, the 6-item pre-delivery checklist, and cross-surface workflow context (companion static-ads-workflow.md for ad recreation; higgsfield-marketing-studio cross-surface-workflow.md §3 for ms_image / DTC Ads Higgsfield-native alternative).

Controls facial expressions in Seedance 2.0 with FACS (Facial Action Coding System) Action Unit codes — muscle-level direction (AU12 = lip-corner puller, AU6 = cheek raiser) instead of emotion labels. Use whenever the user wants precise facial acting, a forced/uncanny/mixed expression, micro-performance in a close-up, monologue or dialogue facial beats, a 'which AU code for anger/fear/disgust' answer, or to generate a FACS reference sheet for a character. Pairs with higgsfield-soul Micro-Expressions (named expressions), higgsfield-audio (dialogue + lip-sync), and higgsfield-gpt-image-2 (the reference-sheet image).

Use when the user wants to run a full ad-campaign pipeline on top of Higgsfield Marketing Studio — 'create a campaign', 'build a content plan', 'run the content pipeline', 'generate 100 UGC videos', 'plan and schedule a launch', 'make a batch of ads from my product', or 'how much did this save vs traditional production'. Covers the 5-stage orchestration (Research → Plan → Generate → Publish → Report), the UGC-first 5-format campaign mix (UGC Entertainment, Street Interview, Unboxing, Product Review, ASMR), the even-split allocation math, button-driven onboarding, the per-batch generation gate, and the publish + cost-report tail (satellite). Defers all Marketing Studio API ground-truth (presets, params, hooks, avatars) to higgsfield-marketing-studio.

Guides users through professional filmmaking workflows in Higgsfield Cinema Studio, including creating multi-shot sequences, configuring optical stacks, applying color grading, managing Soul Cast AI actors, and structuring per-scene prompts with Director Panel camera movements. Use when the user mentions Cinema Studio, Cinema Studio 2.5, Cinema Studio 3.0, Cinema Studio 3.5, Cinema Studio 4.0 (`cinematic_studio_video_4_0` — its four modes, camera/lens/aperture/era/genre/pacing ids, light presets, color palette), Soul Cast, color grading, multi-shot video, shot sequences, storyboard workflow, Hero Frame, optical stack, keyframe interpolation, Elements system (@Characters/@Locations/@Props), Speed Ramp, Director Panel, Higgsfield Popcorn, Single Shot / Multi-Shot Auto / Multi-Shot Manual modes, Reference Anchor, Smart shot control, or any professional filmmaking workflow inside Higgsfield.

Pre-production story-and-character development for Higgsfield projects — the upstream layer that decides WHAT to prompt before any model runs. Use when the user wants to build a character, design a world, develop a story or premise, create a character sheet / character bible / story bible, lock a visual style or 'visual DNA', plan a multi-shot narrative with consistent characters, or says things like 'help me design a character', 'build the world', 'I need a backstory', 'make this character consistent across shots', 'develop my film/series concept', or 'I keep getting generic AI characters'. Routes the locked outputs into higgsfield-prompt + the right model. Adapted from Higgsfield's official character-design framework by @vavavinca.

Use when the user mentions Higgsfield Canvas, a node-based or node graph workspace, an infinite board/canvas, chaining generations into a pipeline, or wants to wire prompts → images → videos across models on one surface. Covers what Canvas is, the node categories, the seven models that run inside Canvas, the named canvas patterns (Simple Seedance, Extend Video, Image Edit, StoryBoard With Elements, Long Video fan-out), the build-free / generate-paid cost model, reusable templates, assets-as-nodes, and Shared Canvas live collaboration. Also trigger on 'Higgsfield ComfyUI alternative', 'node workflow', or 'connect nodes to build a scene/campaign'.

Use when the user asks about camera movements, shot types, or how to describe camera behavior in a Higgsfield prompt. Contains all named camera controls with descriptions, best use cases, and example prompt phrases.

Use when the user asks about audio in Higgsfield videos, needs to add dialogue or lip-sync, wants sound effects or ambient sound in generated video, asks about music or BGM in output, or is using any audio-capable model (Kling 3.0, Seedance 1.5 Pro, Seedance 2.0, Veo 3/3.1, Grok Video / Grok Imagine). Also use when the user's prompt would benefit from audio direction but they haven't mentioned it. Also use when the user wants standalone audio — a soundtrack, ambience bed, multi-speaker scene audio (Seed Audio 1.0), or text-to-speech voiceover. Also use to swap or revoice the speaker in an existing video (voice_change), or to clone / create a reusable voice (create_voice → a voice_type 'element' voice usable in TTS and voice change).

Use when the user asks about Higgsfield Assist (the built-in GPT-5 copilot), how to use the platform's native AI assistant, credit optimization strategies, plan selection, how to get more from fewer credits, or platform efficiency tips.

Use when the user asks about Higgsfield's one-click Apps, wants to know which app to use for a specific output, or needs guidance on the Apps workflow.

Writes the character-performance layer of a video prompt as behavior under pressure, not displayed emotion — objective, obstacle, tactics, beats, subtext, listening, body/status/proxemics, and mandatory eye life. Produces a reusable 150–220-word acting master profile per character plus a per-scene rewrite of it, and a locked voice prompt. Use whenever a prompt needs acting, performance, or emotion direction; whenever characters read wooden, dead-eyed, or 'AI'; whenever a character must stay themselves across many shots; or when the user asks for character behavior, mannerisms, tics, a gait, or how someone reacts. Pairs with higgsfield-facs (muscle-level AU codes) and higgsfield-seedance (the prompt the paragraph goes into).

Use when the user wants a 3D model, mesh or GLB out of Higgsfield — image-to-3D, multi-view-to-3D, text-to-3D, rigging or animating a mesh, remesh / retopology, retexture, or 3D Body (human shape + pose from one photo) — or mentions 3D Jutsu / Scene Builder 3D (private Blender scene projects built through Python). Also use for a 3D turnaround of a character or prop as a consistent multi-angle reference, or a physical 3D blockout rendered as a staging reference for a video shot. Names that trigger it: generate_3d, sam_3_3d, image_to_3d, multi_image_to_3d, meshy_v7_image_to_3d, tripo_h3_1_image_to_3d, tripo_h3_1_multiview_to_3d, hunyuan3d_v3_image_to_3d, meshy_v6_text_to_3d, hunyuan3d_v3_1_text_to_3d, tripo_3d, 3d_rigging, meshy_v5_remesh, meshy_v5_retexture, sam_3_3d_body, animation_actions, scene_builder_3d_*.

Use this skill whenever the user asks anything about Higgsfield AI — writing or refining video/image prompts, choosing a model (Kling, Veo, Wan, Seedance, Minimax Hailuo, DoP, Soul, Nano Banana, Seedream, Flux, GPT Image, etc.), camera controls, named motion presets, Soul ID character consistency, Cinema Studio 2.5/3.0, Vibe Motion, troubleshooting failed generations, credit optimization, Photodump, or any mention of higgsfield.ai. Also trigger on generic "write me a video prompt" or "make me an AI video prompt" requests when Higgsfield is the user's configured platform.

雪峰式AI-Native产品开发方法论。适用于：(1) 用户行为开放、不可穷举的AI-native产品（AI日历、AI助手、AI推荐、对话式产品等），(2) 强模型依赖型场景，AI驱动核心决策而非仅辅助，(3) 多专精Agent架构设计与分工，(4) 上线后快速校准、行为审计与漂移检测，(5) 模型选择和智能路由策略，(6) 概率性输出的质量评估。触发场景包括"AI-native产品怎么做"、"用户行为不可预测怎么办"、"多agent怎么分工"、"模型漂移怎么处理"、"校准到95%太难了"、"唯快不破"、"怎么选模型"、"agent并行分工"、"AI产品上线后怎么迭代"。注意：如果产品是场景明确、边界可定义的+AI类型，请改用 keqian-method skill。即使用户没有明确说"AI-native"，但在讨论AI驱动决策、用户行为不可预测、概率性输出等话题时也应触发。