AI Command Deck

What this skill does

• 109 defined commands with meaning, when-to-use and when-not, combinations, conflicts and an evidence label
• Manual commands and chains take priority: in the right order, format always last
• Cautious auto-routing: at most two commands; "no routing" is a valid result
• Aliases and typos: /ideas → /brainstorm, /concise → /brief, "Did you mean /critic?"
• Prefers fitting installed skills and tools: only those that really exist in your environment
• Transparency line + "deck why": see what was applied and why
• You're in control: transparency level, auto on/off, skill states; lasting wishes as a Deck profile to paste
• Helper script deck.mjs: parser, catalog, matrix validator and 33 automated tests

Description

AI Command Deck is a skill that turns the viral "ChatGPT commands" into a real, defined command system. Those lists promise secret commands: in reality no AI has them built in; /critic or /summarize only work because the model interprets the word somehow, a little differently every time. AI Command Deck defines for 109 commands what they mean, when they apply and when not, what they combine with and what they conflict with: 100 well-known commands from the lists plus nine of its own such as /brief, /deep, /factcheck and /next. Popular variants (/concise, /verify, /tldr) are recognized as aliases, and typos like /critc get a "Did you mean /critic?". You can type commands manually and chain them (/compare /brief /table), they always take priority, or just ask normally: the deck detects the intent and picks at most two fitting commands; for "Thanks!" or simple questions it deliberately picks nothing. Chains run in the right order (ideas before critique, cause before solution, format last), and conflicts like /shorten and /expand are caught and resolved with a single question. When an installed skill or an available tool clearly does the job better (say, a decision skill), it prefers that, following your settings (preferred, automatic, manual-only, disabled). A compact transparency line shows what was applied, and "deck why" explains the choice. The top rule is capability before claim: no web access means no "verified", no email claimed as sent, no setting claimed as saved, and a draft never turns into an action unasked. Every command carries an evidence label (proven to help / with limits / style only), and /godmode is honestly what it is: an alias for a defined recipe, not a hidden mode. Because many agent tools intercept a slash at the start of a message, the deck is called collision-free (/command-deck … in Claude Code, $command-deck in Codex, @command-deck in Windsurf; in chats, commands go in the text).

Examples

Tested with Claude Code

Illustrative answers. Results may differ depending on the AI and model.

What it does not do

• No secret or official commands: nothing gets "unlocked"; the commands are defined instructions
• No faked capabilities: no claimed web search, code run, skill use or saving that didn't happen
• No actions without a request: writing ≠ sending, planning ≠ creating; no permission bypass
• No heavy orchestration: no command pipelines for simple questions, no editing of your configuration

Compatibility & tech

Tested (internal)

5 scenarios

Recommended runtime

Works with any current model (Claude, GPT, Gemini). The helper script needs Node (no extra packages) and is optional, in chats the deck runs purely on its instructions.

Modes

Manual commands and chains · Auto-routing (cautious) · Controls: deck help, commands, why, prefs, reset · Recipes (decision, problem solver, deep review, fact check, content improve)

Inputs

a normal request (auto-routing) · slash commands, single or chained · control words after the deck call (deck why, deck prefs …)

Output format

The answer to your request with a compact transparency line above it (e.g. ⚡ /compare · /risks · /decision), optionally detailed or hidden. For conflicts, one short question; for missing capabilities, an honest one-line note.

Subcategory

AI slash commands, prompt shortcuts & skill routing

License

Proprietär

Security profile

What you get

• command-deck-1.0.1/23 files
  • .claude-plugin/marketplace.json
  • plugins/command-deck/skills/command-deck/23 files

    • SKILL.md
    • manifest.json
    • data/4 files

      • command-matrix.json
      • tests/3 files

        • benchmark-results.md
        • benchmark.json
        • deterministic.json
    • references/16 files

      • capabilities.md
      • commands-build.md
      • commands-create.md
      • commands-decide.md
      • commands-learn.md
      • commands-marketing.md
      • commands-meta.md
      • commands-output.md
      • commands-plan.md
      • commands-social.md
      • commands-think.md
      • commands-verify.md
      • commands-write.md
      • platforms.md
      • preferences-security.md
      • routing.md
    • scripts/1 file

      • deck.mjs
  • .agents/skills/command-deck/→ universal, same content (Codex, Cursor, Copilot, Gemini, Windsurf, Cline)
  • LICENSE.txt

Installation

Also works as a chat prompt

Note

The commands are not official features of ChatGPT, Claude or any other provider: they are defined instructions of this skill. Settings apply to the session; you paste lasting wishes yourself as a Deck profile. In some tools the deck has to be called first because they intercept a slash at the start of a message. The helper script needs Node (no extra packages).

Changelog

• v1.0.124.09.2026Maintenance update: follow-up questions now use your platform's native question tool (at most 3 questions, recommended option first, numbered list as fallback, never blocks unattended runs). Plus cleaner typography across all shipped files.
• v1.0.023.09.2026Initial release: gives popular AI slash commands a fixed meaning and picks the smallest useful combination of commands, installed skills and available tools for each request, transparent and under your control. 109 defined commands (100 well-known + 9 new) with aliases, typo detection, chains, conflict rules, evidence labels and recipes; cautious auto-routing (max 2 commands, no routing is a valid result), transparency line + 'deck why', session preferences + Deck profile. No faked capabilities, no actions without a request. Collision-free invocation per tool. Helper script deck.mjs (parser, catalog, validator, 33 tests), command matrix as data source, 4 rule + 12 command references.

Frequently asked questions