0xAddict

fable-harness

Run an autonomous Opus-Generator + Fable-Verifier harness against a sprint contract with hard ≥98% spec gate, file-based handoff, cron auto-unstick, Monitor on status.txt, read-only approved contracts, session-scoped Stop gate, optional Codex red-team adversary (prompted to BREAK the build), and CronDelete-self on PASS. Use when the user says "fable harness", "fable-harness", "opus builder fable verifier", "opus thinking-high generator", "sprint contract", "spec gate", "hard spec gate", "TDD-driven harness", "Generator + Verifier", "auto-unstick", "go end to end don't ask", "harness this", or "multi-task autonomous build" AND wants the Opus-thinking-high builder / Fable verifier split (use the base `harness-contract` skill instead when the user wants the default Sonnet builder / Opus verifier). Triggers strongly when the request bundles ≥3 deliverables, wants measurable must_pass criteria, no further questions, a hostile Codex red-team gate, and self-termination when the contract is met.

harness-contract

Run an autonomous Generator + Verifier harness against a sprint contract with hard ≥98% spec gate, file-based handoff, cron auto-unstick, Monitor on status.txt, read-only approved contracts, session-scoped Stop gate, optional Codex adversarial verification, and CronDelete-self on PASS. Use when the user says "sprint contract", "spec gate", "hard spec gate", "TDD-driven harness", "Generator + Verifier", "auto-unstick", "go end to end don't ask", "harness this", or "multi-task autonomous build". Triggers strongly when the request bundles ≥3 deliverables and wants measurable must_pass criteria, no further questions, and self-termination when the contract is met.

harness

Launch the Generator-Verifier harness for long-running app development. Spawns Opus Generator and Sonnet Verifier with sprint contracts, file-based handoff, and /loop 1m auto-unsticking monitor.

workctl

管理 Work Agent 平台能力。通过 `workctl schema` 发现动态产品和命令，再用结构化输出执行操作。

workctl-operator

安装、升级、认证、发现并使用 Work Agent CLI (`workctl`)。适用于通过 `workctl schema` 发现并调用阿里巴巴国际站商家经营工具，处理店铺经营数据、广告、发品、商品优化、AI 图片/视频、旺铺、选品、物流、交易、IM、知识问答、深度研究等场景。也适用于配置 workctl、登录认证、排查 token/cache/schema/runtime/async task/recovery 问题，编排多个 workctl 工具，或为新业务域接入 registry/discovery/detail/tools/call 协议。执行业务命令时必须以当前 `workctl schema` 和 `workctl <command> --help` 为准，不照搬旧 MCP/CLI 命令。

agent-taskboard-conventions

Conventions for reading and writing the Agent Taskboard (Trello board id 69fb292ef38b61c3bb1be54d, accessed via the tasks-MCP). Use BEFORE any card_create / card_update on the Agent Taskboard. Covers lists/stages, the 4-dimension card encoding (brand label + action label + stage list + lane owner), brand label table, lane roster, approval flow, auto-rules, and the agent onboarding contract. Triggers when an agent is about to create, move, label, or edit a card on the Agent Taskboard.

elevenlabs-tts

Generate natural-sounding text-to-speech audio using the ElevenLabs API. Use this skill whenever you need to produce a voice/audio reply instead of text — especially when a Telegram user sends a voice message asking for audio back ("can you reply via audio", "I'm driving and can't read"), when the user explicitly requests audio output, when you're making a spoken summary/briefing, or when you're tempted to fall back to macOS `say` (don't — it sounds robotic). Also use this when packaging a spoken message to a human who will listen rather than read. ALWAYS prefer this over `say` unless the user explicitly asks for the built-in voice.

evidence-investigate

Use when answering questions about the CURRENT state of code, git history, or on-disk files — e.g. "is X installed", "what does Y do", "what branch has Z", "when did W change", "does the running code have feature F". Do NOT use for trivial file existence checks answerable by a single ls, for implementation/edit work, or for questions about mutable external state (APIs, databases, live processes, Telegram history).

mcp-restart-ceremony

Reconnect threadwork agents (boss/steve/sadie/kiera) to the task-board MCP server after disconnects, /clear ceremonies, or team-wide reconnect requests. Reads pane first; ESC-ESC clears; opens /mcp picker; verifies task-board ✔ connected · 40 tools. Surfaces field findings F0-F4. Do NOT run on snoopy.

nike

Apply the LOCKED Nike design system / "Nike taste" to any creative — kinetic-typography video overlays (Remotion), static graphics, social posts, and ads — by re-applying the exact proven design tokens verbatim (Anton heavy condensed UPPERCASE, stark monochrome white-on-near-black, exactly ONE Volt #D7FF1E accent, square hard edges, snappy 7f-in/6f-out motion, non-overlapping captions) and regenerating fresh on-brand Higgsfield accent imagery each run. Use this skill whenever the user says "/nike", "nike style", "nike taste", "nike design", "apply the nike tokens", "make it look like nike", "Just Do It style", "Nike kinetic typography", or asks for athletic high-contrast condensed-caps creative with a single volt-green accent — even if they don't name the brand explicitly. Trigger it before designing any overlay, ad, or graphic that should match this house style so the tokens are applied exactly, not paraphrased.

premortem

Run a premortem on a plan, launch, product, hire, strategy, partnership, pricing change, or high-stakes decision by assuming it failed 6 months from now and working backward to identify why. Use when the user says "premortem this", "premortem my", "run a premortem", "what could kill this", "future-proof this", "stress test this plan", "what am I missing here", "find the blind spots", "what could go wrong", "am I missing anything", "poke holes in this", "where will this break", or "devil's advocate this" in relation to a concrete plan or commitment. Do not use for simple factual questions, ordinary draft feedback, vague ideas with no plan yet, or LLM Council requests.

recycle

Evacuate-then-rehydrate context recycle. Run when context-budget-watch warns at 60%, when the user manually requests recycle, or proactively before drift compounds. Captures session state to durable memory and injects /clear.

structured-image-gen

Edit and generate images using structured JSON prompts with Nano Banana 2 (Gemini image model). Use this skill whenever the user wants to edit AI images, change colors/materials/objects in photos, swap logos or text, adjust lighting or camera perspective, generate product images, or work with Nano Banana / NB2 / Gemini image generation. Also trigger when the user mentions JSON prompts for images, reference folders for image editing, or wants fine-grained control over image modifications without the AI ruining the rest of the image.

task-delegation

Use when receiving any task from the task board, before doing any direct work — enforces delegation to persistent runner or sub-agent, with per-task monitoring loops. Main thread is a dispatcher, never a worker.

voice-transcribe

Transcribe voice messages and audio files using ffmpeg and OpenAI Whisper. Use whenever a Telegram voice message arrives (attachment_kind="voice", audio/ogg), when the user sends an audio file, asks to transcribe audio, or when you receive any .oga/.ogg/.mp3/.m4a/.wav file containing speech. Automatically triggers on inbound Telegram voice messages.

grill-me

Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".