prompt-optimizer

Prompt Optimizer

Analyze a draft prompt, critique it, match it to available components, and output a complete optimized prompt the user can paste and run.

Detection tables, component-matching tables, the full output-format spec, and worked examples live in reference.md in this skill's directory — read it before producing output. It lists herow components; the installed plugin set varies, so verify a component exists in the current session before recommending it.

When to Use

• "Optimize this prompt" / "improve my prompt" / "rewrite this prompt"
• "Help me write a better prompt for…" / "what's the best way to ask Claude Code to…"
• User pastes a draft prompt and asks for feedback or enhancement

Do Not Use When

• User wants the task done directly (just execute it) or says "just do it"
• "Optimize this code/performance" — refactoring, not prompt optimization
• User is asking about tooling setup rather than prompt content

How It Works

Advisory only — do not execute the user's task. Do NOT write code, create files, run commands, or take any implementation action. Your ONLY output is an analysis plus an optimized prompt. If the user asks for execution, tell them to make a normal task request instead.

Run this pipeline sequentially, then present results in the Output Format below.

1. Phase 0 — Project detection. Read CLAUDE.md if present; detect the tech stack from project files (signal table in reference.md). No project files → flag "tech stack unknown".
2. Phase 1 — Intent detection. Classify into: New Feature, Bug Fix, Refactor, Research, Testing, Review, Documentation, Infrastructure, Design (signals in reference.md).
3. Phase 2 — Scope assessment. TRIVIAL (single file, <50 lines) → direct execution · LOW (single module) → one command/skill · MEDIUM (multi-component, same domain) → command chain + verify · HIGH (cross-domain, 5+ files) → plan first, phased execution · EPIC (multi-session/PR) → blueprint skill.
4. Phase 3 — Component matching. Map intent + scope + stack to concrete commands/skills/agents using the tables in reference.md, validated against what actually exists in the session.
5. Phase 4 — Missing-context detection. Check: tech stack, target scope, acceptance criteria, error handling, security, testing expectations, performance constraints, UI/UX (if frontend), database changes (if data layer), existing patterns, scope boundaries. If 3+ critical items are missing, ask up to 3 clarification questions first and fold the answers in.
6. Phase 5 — Workflow & model recommendation. Place the prompt in Research → Plan → Implement (TDD) → Review → Verify → Commit; recommend a model tier and, for HIGH/EPIC, a multi-prompt split (tables in reference.md).

Output Format

Respond in the same language as the user's input, with these sections (full spec and examples in reference.md):

1. Prompt Diagnosis — strengths, issues table (issue/impact/fix), open questions
2. Recommended Components — type/component/purpose table incl. model
3. Optimized Prompt — Full Version — one self-contained fenced block: task + context, tech stack, command invocations at the right stages, acceptance criteria, verification steps, scope boundaries (what NOT to do)
4. Optimized Prompt — Quick Version — compact one-liner pattern for the intent
5. Enhancement Rationale — what was added and why
6. Footer — "Not what you need? Tell me what to adjust, or make a normal task request if you want execution instead."