Explore

Create and inspect science projects in the autoresearch-mlx Harness workspace, including its local starter and optional upstream integration.

Details behind AGENTS.md's 'Starting a model on this computer' flow: reading `fleet models`, sizing a start, choosing file and engine, picking a model for an engine with none, what `fleet verify` checks, and where to read when unsure. Open it when a step of that flow needs more than the flow says.

Look after the models on the user's machines — one laptop or a fleet: recognise their private grid, pick the model and settings that fit this machine for coding without asking, start it with the longest context the machine can give (never under 64K), say what it costs in memory, prove it answers with one bounded call, change or stop a running one, and use Grid routing, usage, media and training commands.

Serve a Hugging Face model with vLLM on a Linux machine with an NVIDIA or AMD GPU, configured from the model's official vLLM recipe — or, when it has none, from the model's own files — and join it to the person's fleet. Load before installing, configuring, starting or stopping vLLM.

Serve a Hugging Face model with SGLang on a Linux machine with an NVIDIA or AMD GPU, configured from the model's SGLang cookbook page — or, when it has none, from the model's own files — and join it to the person's fleet. Load before installing, configuring, starting or stopping SGLang.

Reuse what Ollama already has on this computer: adopt a running Ollama server into the person's fleet, or serve an Ollama-downloaded model without Ollama. Load before touching Ollama, its models or its settings.

Serve an MLX or Hugging Face safetensors model already on this Mac with mlx-lm's server and join it to the person's fleet. Load before installing, starting or stopping mlx-lm, or when `fleet models` lists an `mlx` or `safetensors` model on Apple silicon.

Reuse what LM Studio already has on this computer through its `lms` CLI: adopt its running server into the person's fleet, serve its MLX models, or serve its GGUF files without it. Load before touching LM Studio, its models or its server.

Create and inspect music projects in the Ableton AI Harness workspace, including its local starter and optional upstream integration.

Harness runs coding agents (Claude Code, Codex and others) as harnesses, sessions the person can watch and join in the Harness app, on this computer or their other machines. Use it to start another agent on work that can run beside yours, needs another agent's view (a review, a second opinion from Codex), takes long, or belongs in its own folder; then wait for it, read its answer and close it. Do small, quick work yourself, and quick fan-out inside one task with your own sub-agents.

ALWAYS invoke this skill when the user asks what to do next, what is left, or what is blocked - "what's next", "what now", "what should we work on", "anything I can do" - in any language. This skill picks the next piece of work; when the user asks HOW to do a thing or says they do not understand what to do, that is os-step-by-step. Reads the last report, local changes, open pull requests and always the backlog. First finishes what is finished: verified-ready pull requests merge in the same pass. Then sorts the rest into doable-alone and needs-you, ending with one recommended next task in plain words - what it closes or unblocks. Never invents tasks.

ALWAYS invoke this skill when you need the user to act - run a command, paste a secret, click, approve - and whenever they ask how to do something or say they do not know what to do: "step by step", "walk me through it", "what do I do", "what should I do", "I don't understand what to do", "explain what I need to do", in any language. Picking which task comes next is os-whats-next; this skill is for doing the thing in front of you. First earn the ask: try it yourself, find another route, shrink it to the part only they can do. Then one action per step, commands labelled by what they touch, no jargon. Commands are single lines that prompt for any value - typing hidden for secrets - and confirm in plain words. Afterwards verify their step.

ALWAYS invoke this skill when the user asks for simpler or shorter about something said or written - "say it simply", "what does this mean", "I don't understand your answer", "too long", "wait, what?", "bro" - in any language, about any text: your own answer, a report, a review comment, an error. "I don't understand what to DO" is os-step-by-step; this skill restates text. It restates for a reader who does not read code: leads with the point, keeps every number, warning and caveat - no facts added, no bad news dropped. A number returns exactly that many points, most important first.

ALWAYS invoke this skill when work wraps up or the user asks how it went - "done or not", "are we done", "what happened", "report", "what's the status of this ticket" - in any language, and when a Stop hook asks for a session report. Produces a ten-line plain- language report: a lead, a checkmark table, and a verdict - fully done, anything needed from you, new debt, safe to close. Every "yes" names its proof; unverified says "not checked". Saves the report so the next session starts from it instead of re-exploring the repo.

ALWAYS invoke this skill when the user asks about work done outside this session - "check work", "check the others", "check other sessions" - or to accept one: "the session is done, check it", "can we merge it" - in any language. You are the receiving party: treat the report as a claim, verify each part against machine state, name every gap between claimed and true. Verified-ready work merges in the same pass. End with what is left, one next step, and any reply another session needs, ready to paste.

ALWAYS invoke this skill before asking the user any technical question or offering options, and whenever they ask to be asked in plain words - "ask simple", "ask me simply", "ask me in plain words" - in any language. ALWAYS invoke it too when the user asks for your view on a technical choice: whether it is worth doing, more than the problem needs, or replaceable by something simpler, and which option you would take. Invoke it even when the code makes the answer look obvious: the checks are what make the answer more than a guess. Rewrites the question in plain words and always ends with one marked recommendation. A structural choice first passes six checks, shown as a table: effort now, simpler substitute, extra work later, lock-in, over-engineering, easy to undo. Doing nothing is always weighed.

Review generated or changed WordPress code — plugins, themes, and blocks — before it ships. Best used reactively after an agent writes, edits, or reviews code touching WordPress APIs: add_action/add_filter, shortcodes, meta boxes, AJAX handlers, REST routes, WP_Query or $wpdb, widgets, or WP-CLI commands. Use on 'review this plugin', 'is this safe to ship', 'make this translatable', 'speed up this query', or after tasks like 'write a plugin' or 'add an endpoint/shortcode/meta box'. Enforces escaping and sanitization, nonces plus capability checks, prepared database queries, core-API-first development, translation-ready strings, and query/caching discipline. DO NOT USE for WooCommerce-specific order, product, or checkout logic (use woo-guard), non-WordPress PHP, generic code quality review (use clean-code-guard), test code review (use test-guard), server or hosting configuration, or conceptual WordPress questions.

Review generated or changed WooCommerce code — extensions, payment and shipping integrations, checkout customizations, and order/product logic — before it ships. Best used reactively after an agent writes, edits, or reviews code touching WooCommerce APIs: wc_get_order, wc_get_orders, wc_get_product, WC() cart or session, woocommerce_* hooks, Store API endpoints, payment gateways, order or product meta, HPOS, subscriptions, or bookings. Use on 'review this Woo plugin', 'is this HPOS compatible', or after tasks like 'write a WooCommerce extension', 'add a checkout field', 'hook into the order flow', or 'update stock'. Enforces HPOS-safe order access, CRUD over direct meta, feature-compatibility declarations, server-side checkout validation, money-handling discipline, and hooks over template overrides. DO NOT USE for WordPress code without WooCommerce APIs (use wp-guard), generic code review (use clean-code-guard), test review (use test-guard), or store configuration and admin-screen questions.

Review generated or changed test code against universal testing rules before it ships. Best used reactively after an agent writes, edits, generates, or refactors tests, before presenting, committing, or merging them. Use for pytest (test_*.py, *_test.py), PHPUnit/Pest (*Test.php), Jest/Vitest (*.test.ts, *.spec.js), Go (*_test.go), files under tests/, __tests__/, or spec/, and review requests like 'write tests for X', 'add tests', 'test this', 'review these tests', or PR diffs containing tests. Can also guide test writing when explicitly invoked before the work. This skill is the quality gate that prevents AI-generated test bloat. DO NOT USE for production or implementation code review (use clean-code-guard), CI or test-runner configuration, running or debugging tests, or general architecture discussion.

Review generated or changed documentation before it ships — READMEs, API references, docstrings, PHPDoc/JSDoc, changelogs, tutorials, and doc sites. Best used reactively after an agent writes or edits docs, after code changes documented behavior, or before publishing docs. Use when the user says 'review the docs', 'is this documentation accurate', 'update the docs', 'write a README', 'document this API', 'add a docstring', or 'add a changelog entry'. Core job: verify every referenced function, flag, endpoint, config key, and code sample against the source; catch docs-vs-code drift; strip filler and unverifiable claims. DO NOT USE for production code review (use clean-code-guard), test review (use test-guard), marketing copy or blog posts, prose style editing of non-technical writing, or documentation site theming.

Review generated or changed production code before it ships, using Clean Code, SOLID, DRY, KISS, YAGNI, and LLM-specific failure-mode checks in any programming language. Best used reactively after an agent writes, edits, refactors, or fixes code, before presenting, committing, or merging the result. Use when the user asks "review this PR", "is this safe to merge?", "make this cleaner", "audit this code", "refactor this", "fix this bug", or after a coding agent produced implementation code. Can also guide writing when explicitly invoked before a risky edit. Invoke it on your own initiative the moment you finish writing, editing, or refactoring non-trivial production code, before presenting or committing — don't wait to be asked. DO NOT USE for factual/conceptual questions, CI/tooling config, git workflow, running/debugging tests, pure architecture discussion, prose writing, data analysis, or test-code review (use test-guard).

Local-first long-term memory for AI agents via MCP — zero LLM calls at write time, typed retrieval at read time. Use when session decisions and pitfalls should survive across days, when multiple agents or dialogs share facts on the same filesystem, or when you want drift detection so past mistakes don't repeat. Verified claims only — every published number ships as a byte-recomputable, ed25519-signed evidence pack.

Use when the user is about to rely on prior session memory for a non-trivial decision, or when something feels off about a recall hit — runs the Merkle chain check, drift summary, and daemon liveness in one pass and surfaces any tampering, drift bursts, or daemon-down state before the user acts on stale or corrupted memory.

Use only when the user explicitly invokes `$improve-harness`. Test one authorized improvement to repository guidance or capabilities against an observed baseline and a fresh rerun. Do not infer permission from ordinary work or automatically reflect after tasks.

Use only when the user explicitly invokes `$engineering-wisdom`. Provide repository-grounded engineering advice using contextual heuristics. Advice does not establish repository policy or authorize an architecture rewrite.

Use only when the user explicitly invokes `$encode-invariant`. Convert accepted repository rules into native mechanical validation with positive and negative proof. Do not infer policy from conventions, code patterns, tests, defaults, or undocumented preferences.

Reconcile an existing Fizz harness with a changed source tree. Detects added/removed/changed contract functions, quarantines stale properties, regenerates drifted handler stubs, and refreshes the snapshot. Trigger on "fizz-sync", "resync fuzzing", "sync fuzz harness", "refresh fuzzing properties", "fuzzing drift check".

Convert English-language properties in PROPERTIES.md (produced by the Fizz skill) into Solidity assertions inside the existing fuzz harness, then flip their checkboxes. Trigger on "fizz-convert", "convert properties", "implement properties from PROPERTIES.md", "convert PROPERTIES.md to Solidity".

Finds high-value SEO and GEO keywords using web search, AI analysis, and optionally paid tools like Ahrefs or Semrush. Produces a validated keywords.csv file with a fixed schema for downstream pipeline consumption.

Audits a live website for AI-engine discoverability (AEO/GEO). Crawls the site, runs 16 deterministic checks plus a 6-dimension content evaluation, and produces a scored report (A-F) with prioritized fixes. Use to get a baseline before improve-aeo-geo, or to measure progress after changes.

Greet the workspace. Use when the user asks the Starter harness to say hello or to prove the skill is loaded.

Build and verify interactive HTML experiences in Web Studio's live, sandboxed preview.

Turn a user's brief into an editable voxel environment or asset kit in Voxel Worlds. Use for original scene design, modular assets, terrain sculpting, walkable spaces, importing VOX assets and delivering GLB/VOX files with an offline studio.

Find, evaluate, and download common purchasable CAD parts from step.parts, including named off-the-shelf actuators, servos, motors, electronics boards, connectors, screws, bolts, nuts, washers, bearings, standoffs, and other catalog components. Use when Codex needs to search the hosted step.parts catalog before creating simplified placeholder geometry, resolve fuzzy part names, standards, aliases, or dimensions, choose a matching part, fetch a canonical .step file, verify checksums, or use the step.parts API/OpenAPI/catalog endpoints for standard part discovery.

Measure mesh files against Design for Additive Manufacturing (DfAM) rules and report printability findings per process (FDM, SLS, SLA/DLP, metal PBF, MJF). Use when the user asks whether a part is printable, wants overhang/wall-thickness/support analysis of an `.stl`, `.obj`, `.ply`, or `.3mf` mesh, wants a build-orientation recommendation, or wants DfAM redesign guidance before slicing with `$gcode` or regenerating geometry with `$cad`.

Dry-run, upload, and cautiously initiate local Bambu Lab print jobs from validated plain `.gcode`, using Bambu LAN FTPS/MQTT handoffs.

Build coordinated editable DOCX reports, formula-driven XLSX workbooks and fresh LibreOffice PDF previews from structured source. Use for Sheet & Docs Studio work.

Build, explain and verify small quantum circuits in Quantum Studio's editable statevector simulator.

Turn a user's STL and explicit machine/material requirements into compared OrcaSlicer plans, editable 3MF projects, checked G-code and a portable handoff. Use for Orca Slicer harness work.

Turn measured requirements into editable OpenSCAD part families, compare fit variants, and deliver checked STLs plus a portable project.

Plan real experiments and analyze supplied continuous measurements in Lab Bench. Use for defining factors and independent runs, randomized collection sheets, measurement CSV imports, experimental uncertainty and diagnostics, comparisons and follow-up confirmation runs with a reproducible report.

Build Godot games with GDScript, native gameplay assertions and sandboxed Web exports. Use for Godot Studio project and game-mechanic tasks.

Build and verify interactive GeoJSON maps in Atlas, with local exploration, safe imports, feature inspection and export.

Create original vector artwork and complete visual asset sets from a person's brief, with custom drawing programs, user-editable controls and real SVG/PNG/source delivery.

Create and revise original board or card games from a person's idea, with editable mechanics and components, real human playtests, reproducible simulations and a complete playable and printable kit. Use for tabletop game design, card drafting, cooperative puzzles, push-your-luck systems and new rules beyond the example game.

Turn measured design briefs or existing parts into editable FreeCAD projects with checked dimensions, bores, clearances, interference and fastener access; deliver separate STEP/STL parts, reference drawings and a portable project handoff.

Build PlatformIO firmware and inspect real compilation, memory, artifacts and pin plans without implicitly accessing hardware.

Create and revise field-survey plans from supplied GeoJSON boundaries, exclusions, camera geometry and capture requirements. Use Drone Pilot for editable routes and sorties, recorded-flight CSV analysis, and portable GIS, CSV and report delivery.

Turn user CSV files and a question into a typed, joined, repeatable local SQLite analysis with traceable records, browser revisions and verified portable exports.

Create and revise a usable brand identity and coordinated launch materials from a business brief, using the Forme project and the person's own artwork, text and production requirements.

ALWAYS invoke this skill before anything hard to undo gets agreed to - a contract, a purchase, a migration, a launch, a price change, a reorganisation - and whenever the user asks "what could go wrong", "what are we missing", "poke holes in this", or for a premortem or a red team, in any language. Assumes the decision already failed and works backwards to find out why, in a fresh agent that had no hand in making it. Sweeps nine areas and shows what each produced, including the empty ones. Ends on one verdict: go ahead, go but fix these first, try it small first, think again, do not do this.

ALWAYS invoke this skill when the user asks where the project as a whole stands - "where are we", "what's the big picture", "what have we built", "what is in this project", "map the project", "what does this thing even do", "what is stale", "what can we delete", "update the roadmap" - in any language, and whenever a session report has just been written. It keeps one file, BIG-PICTURE.md: what the product is, every feature with how far it got, which parts nobody has touched, and what is queued next. Age and wiring are measured from git; how far a feature got comes from the reports or says "not checked". Where a tracker is connected it offers to open the queue as tickets, after a yes and never before. Never invents work, never deletes code.

29 production-ready scripts for iOS app testing, building, and automation. Provides semantic UI navigation, build automation, accessibility testing, and simulator lifecycle management. Optimized for AI agents with minimal token output.

Set up, review, or repair a CCManager config — `.ccmanager.json` at a git repository root, or the global `~/.config/ccmanager/config.json`. Use when someone wants ccmanager to launch a different agent CLI (codex, gemini, cursor-agent, copilot…), add command presets, run a command on session state changes or worktree creation, auto-generate worktree directory paths, change merge/rebase arguments, rebind the return-to-menu key, turn on auto-approval — or when a ccmanager setting "does not seem to do anything".

Use only when the user explicitly invokes `$onboard-repository`. Inspect an unfamiliar repository, trace one operational path, identify missing capabilities, and propose evidence-backed knowledge improvements. The first pass is read-only; apply exact approved items only within the requested scope.

Use only when the user explicitly invokes `$audit-onboarding-proposal`. Independently audit an onboarding transcript and exact evidence-backed proposals. Remain read-only and verify claims against pinned repository evidence rather than trusting the producer.

Generate Echidna/Medusa-compatible Solidity fuzz suites from Foundry or Hardhat projects. Trigger on "fizz", "generate fuzz suite", "build fuzz harness", "stateful fuzzing", "fuzzing harness", "property testing", and "invariant suite".

知乎视频 → 下载 → 转录 → 存为 Markdown。 支持知乎视频链接、问答内嵌视频和专栏文章，无需登录。

YouTube 视频 → 下载 → 转录 → 翻译 → 存为 Markdown。 支持中英文，英文内容自动翻译成中文，输出中英对照。

X(Twitter) 推文采集 → 统一 frontmatter Markdown。自动区分图文与视频： 图文下载图片本地嵌入，视频提取直链转成文字稿。无需登录（走官方嵌入端点）。