Tested by SkillHammer Chat custom instructions + Agent Skill for Claude Code and Cursor

Prompt Expander

Turns task-like asks into a clearer brief before your AI acts, without pasting a template.

Most AI chats treat a short ask as the whole brief. Prompt Expander makes the model silently expand a task-like message into goal, constraints, success check and out of scope, then act on that brief.

If the rewrite would change the goal, scope or constraints, it shows the brief and waits for OK. Pure chat, yes/no, short confirmations and "just do it" skip expansion.

Works as paste-ready custom instructions for ChatGPT, Claude.ai and Grok, and as an Agent Skill for Claude Code and Cursor. The ChatGPT Free paste is trimmed to fit the 1500-character respond field in one paste.

What's in the download

• chat/chatgpt.md: paste-ready custom instructions for ChatGPT (Free: ≤1500 chars in the respond field)
• chat/claude.md: paste-ready instructions for Claude.ai (profile or project)
• chat/grok.md: shorter paste for Grok custom instructions
• skills/prompt-expander/SKILL.md: Agent Skill for Claude Code and Cursor
• INSTALL.md, tests/ (fixtures + structure checker), TEST-REPORT.md, LICENSE.txt, REFUND.txt

Who it's for

Anyone who uses ChatGPT, Claude.ai, Grok, Claude Code or Cursor for real tasks and wants clearer briefs without pasting a template every time.

What's tested

• 11/11 pack structure and behaviour checks from a clean unzip of prompt-expander-v1.0.2-without-report.zip (1 skipped: TEST-REPORT absent in the without-report build); FAILED STEPS: 0
• Required behaviours present in SKILL.md and chat/{chatgpt,claude,grok}.md: silent default, ask-and-wait when the rewrite would change the job, skip for empty/chat/yes-no/just-do-it, honest non-interceptor limit
• ChatGPT Free custom-instructions live smoke on 24 Oct 2026: A silent expand, B ask-back, C skip — all PASS; ChatGPT paste body ≤1500 characters for the Free respond field
• Skill-application simulation matrix on 24 Oct 2026: 22/22 expected modes (silent / ask / skip + edge cases); documented in tests/fixtures/live-matrix-20261024.md — not a live model API score

What isn't tested

• Live expansion quality inside Claude.ai or Grok custom-instruction sessions
• Claude Code or Cursor in-app skill discovery smoke
• Whether a given model obeys ask-and-wait under tool-use pressure
• Hosted-model API scores beyond the documented ChatGPT Free A/B/C smoke cases
• Mobile custom-instruction character limits beyond keeping chat/grok.md shorter and ChatGPT Free ≤1500
• Windows/macOS path quirks for skills folders

Read the full TEST-REPORT

Requirements

Any of: ChatGPT (custom instructions), Claude.ai (profile or project instructions), Grok (custom instructions), Claude Code, or Cursor. No API keys and no installer beyond paste or copy.

How to use it

For chat: open INSTALL.md and paste the matching chat/*.md body into that product's custom instructions. For Claude Code, copy skills/prompt-expander/ to ~/.claude/skills/prompt-expander/ (or .claude/skills/ in a project). For Cursor, copy it to .cursor/skills/prompt-expander/.