Coding & Development

Risponde a una domanda leggendo la memoria, e offre di rimettere dentro la risposta come pagina nuova. Usala quando l'utente chiede "cosa sappiamo di X", "fammi un'analisi di X", "confronta X e Y", o qualsiasi domanda che richieda di mettere insieme più cose che sa già.

Chiude una sessione di lavoro scrivendo su disco un file di passaggio (handoff) che la sessione successiva legge per ripartire esattamente da dove eravamo, al posto di /compact. Usala quando l'utente scrive /handoff, o quando dice frasi come "salvami il punto", "prepara la ripresa per domani", "chiudiamo qui", "riprendiamo domani".

Top-down staging with ManimGX: plan where everything in a scene is and render it as an animated floor-plan map. Use when someone wants to stage or block a scene, plan camera positions, lenses or coverage, place lights, or check the 180° axis and screen direction — from a full shot list or one casual sentence. Covers environment and key objects, character positions, postures, facings and moves, physical light sources and every camera with its field-of-view cone. Asks a short round of scene-specific multiple-choice questions when the brief is thin, shows the layout and plain-language shot descriptions for approval, then renders an MP4 plus an editable JSON scene spec through a standalone Python CLI.

Desenvolve jogos Roblox em Luau com Rojo dentro da VM e publica por Open Cloud em um lugar de TESTE informado pelo usuário.

Place and follow up real outbound phone calls with the Phrok tools (make_call, schedule_call, get_call, list_calls, hangup_call, get_account). Use when the user asks you to call someone, book or change a reservation or appointment, check opening hours or stock by phone, or asks how a call went.

SKILL CAROSELLI CLAUDE — Crea un carosello Instagram per @aurumiles direttamente qui, senza Canva, nello stile delle slide della Lezione 1 (nero, Oswald corsivo + Inter, accento rosso o altro colore, foto vere ritagliate dagli originali, poco testo, 1080x1350). Usala quando Davide chiede un nuovo carosello senza specificare Canva, o dice «fammelo come le slide della lezione». Per i caroselli in Canva usa `caroselli-canva`.

将已确认的软件需求拆成小 PR，协调独立 worktree 并行开发、实现者运行自证、代码和安全审查、最多三轮收敛、组合验证与人类验收。用于开发任务，不为普通讨论自动开始实现。

Review measured results for the product linked in .zooi/project.json - draft the readings that are due, for actions and for bets, with evidence, for the person to confirm in zooi.ai. Use when the person asks for a zooi review or readings, or when /zooi:run reaches readings.

Handle everything zooi.ai says is due for the product linked in .zooi/project.json, in order - interrupted runs, approved work orders, weekly planning, readings. Use when the person asks to run zooi, or when zooi or the session start says a run is due.

Carry out an approved zooi.ai work order for this repository - claim it, make the change on a branch, open a pull request the person merges, and deliver the receipt. Use when the person asks to do zooi work, names a work order, or when /zooi:run reaches approved work.

在用户明确选择 LaTeX，或任务确需公式、参考文献等 LaTeX 排版能力时，创建、修复和检查 .tex 与 PDF。涵盖引擎、包兼容、环境诊断和实际页面验收；普通文档、幻灯片或一般 PDF 导出不自动改用 LaTeX。

审查 PR 中的死代码、无意义抽象、重复实现与可复用替代方案，输出带证据和统一口径的 PR 反馈。仅提出意见，不自动删除、重构、安装依赖或采用外部代码。

审阅或按授权精修现有技术文档，改善重复、标题、信息顺序、图表与示例，保留行为契约、必要警告和可访问性。用于文档太长、难找信息、发布前审阅或明确要求改写；不负责从零创建文档。

从文字、文档或参考图创建和调整可编辑流程图、技术路线或研究路线图。默认交付本地可编辑 SVG，用户指定格式优先；只有明确要求飞书/Lark 发布时才加载平台参考，不默认创建外部文档或发起授权。

Report a skillgram bug or a missing-skill gap with real evidence: pick the channel, capture the JSON, open the issue.

Discard staged evidence captures that were not attached (one by id, or all).

Shared operating protocol for every corp team lead (preloaded into lead agents). How leads plan, staff their swarm, brief workers, verify, and report to the manager.

Use when the user asks about their household's dinners, meal plan, recipes, pantry, shopping list, leftovers, who is eating when, or inviting someone to dinner, and the SupperSecret tools are connected.

Corp Talent (HR / Corporate Assistant) - search GitHub and the web for high-quality Claude Code skills, plugins, and MCP servers for a team or gap, have Security vet the finalists, and present a ranked shortlist for the user to approve. Never installs anything on its own. Use when the user runs /corp:hire, or asks to find, recruit, or scout new skills, plugins, or MCP servers for the corp teams.

Prepare manual implementation and correction prompts for Claude Code without invoking it. Use when the user asks Codex or Cursor to pass work to Claude or returns its result for another manual iteration. Make sure to intercept delegation requests even when the user does not ask for a prompt; never invoke Claude automatically.

Maintain coding-agent instructions and their discovery paths. Use when creating or changing AGENTS.md, CLAUDE.md, agent rules, or instruction projections. Make sure to use this skill whenever agent guidance or its deployment changes, even if the request names only one agent.

Qualify regression tests with AssertLedger. Use when asked to check whether a test detects a declared bug or to replay AssertLedger evidence. Make sure to use this skill whenever that qualification is requested, even if AssertLedger is unnamed. Excludes routine TDD, coverage, general code review, and unsupported test runners.

Systematic QA and bug-hunting for simple 2D or 3D games, especially web and Three.js games. Use when asked to test, playtest, QA, debug, or find bugs in a game, or to verify a game against its RULES.md. Produces a test plan, automated checks where possible, and a severity-ranked bug report.

>- Commits the working tree in several focused commits, each containing only related changes, then pushes to the remote. Use when the user invokes /commit-batches, asks to commit in batches, split commits, or batch commits by topic.