generic-ai-skills

Lead an image-first visual-direction walk and write a durable beauty-profile.md. Use when a user wants a visual direction, style, look, aesthetic, or art direction for any medium, including brands, websites, apps, presentations, documents, campaigns, packaging, photography, video, environments, data visualization, or image generation.

Strategic thinking partner for messy topics that need to be explored, untangled, and shaped before they become a plan or decision. Use when the user says things like "help me think through this," "let's kick this around," or "brainstorm this with me."

Executive briefing partner for quickly understanding complex material. Use when the user provides pasted text, a file, a URL, or a topic and wants a concise brief with bottom line, key judgments, confidence, decisions, and gaps. Do not use for long-form research reports or polished prose drafts.

Route explicit ChatGPT work: $chatgpt, ChatGPT orders, Deep Research, Agent, or Pro review. Decide whether the task stays local or becomes a ChatGPT order, then place it through host-supported browser control or prepare a bounded fallback packet.

Route bounded work to local Claude Code, Codex, Gemini, Grok Build, Cursor, or OpenCode CLI subagents through guarded wrappers. Use when a CLI agent, second-model pass, or supervised fast-model branch would improve a task; Gemini, Cursor, and OpenCode are explicit-only inactive nodes.

Gather high-signal activity from user-configured sources into a daily note with stable deduplication and per-source checkpoints. Use for a best-effort cross-source context sweep, a dry run, or a bounded sweep since a specific time.

Walk through supplied feedback, research, notes, audits, transcripts, plans, drafts, or links one decision at a time; keep a running decision log and finish with an implementation checklist. Use only on explicit invocation or item-by-item walkthrough language; do not use for ordinary summarization or recommendations.

Economist council for AI, labor, productivity, growth, inequality, and regional adaptation questions. Use when the user wants sourced convergence, divergence, and pushback from a standing panel before a talk, decision, strategy move, article, or research memo.

Run guarded Google Workspace command-line operations through local helper scripts. Use for bounded Drive, Sheets, Gmail, or Calendar reads; for approved mutations; or when exact JSON evidence and restore-safe handling matter.

Prepare for a meeting, close it out, find meeting history, or review patterns across meetings using available source material. Use for prep, debriefs, decisions, open loops, or source-backed meeting reviews.

Co-create a day, week, or month plan through lightweight approval phases. Use when the user wants to plan a period, reconcile tasks and calendar constraints, update an existing plan, or decide what belongs now versus later.

Plan and run larger projects through a guided session, blueprint, and portable runner. Use when work is too large or fuzzy for one prompt and needs clear steps, tool routing, status checks, or resume support. Use harden for bounded verify/revise runs on an existing thesis, draft, or plan.

Interview-driven intellectual sparring partner for strengthening and pressure-testing an idea, argument, strategy, or decision. Use when the user wants the strongest version of a view, plus its best counterarguments, assumptions, tradeoffs, risks, and next steps.

Practical coaching skill for moments when the user feels stuck, scattered, avoidant, or out of rhythm. Use to run a one-question-at-a-time reset that surfaces the real tension, identifies the primary blocker, and ends with a small plan for today and tomorrow. Not for crisis or clinical care.

Turn settled discussions or supplied sources into the simplest useful, editable FigJam diagram, with semantic, structural, visual, and compression review before handoff. Use only when the user explicitly invokes `$visualize`; do not trigger for ordinary explanation, planning, analysis, or requests that merely mention diagrams.

Explicitly close out the active work thread into configured local records. Use only when the user asks to log, close out, or record the current thread; capture outcomes, decisions, deadlines, dependencies, blockers, and next actions without broadening into unrelated systems.