dhpk-prompt-optimizelisted

Rewrite a raw task prompt into a model-aware, effort-calibrated version before you run it. Use when: the user asks to "optimize this prompt", "improve this prompt for Claude", "what effort should I use", "make this a good API prompt/template", or pastes a rough instruction before a big Claude Code or API task. Detects the target model (current session or a named one), classifies task complexity, recommends an effort level (low/medium/high/xhigh/max, plus the dhpk agent-frontmatter/Claude Code equivalent), asks up to 4 AskUserQuestion questions only for missing required info, then applies model-specific behavioral rewrites. Not for: generic few-shot/CoT/template technique coaching with no model or effort selection (use prompt-engineering-patterns), or auditing token/cache budget (use dhpk-harness-budget). Output: optimized prompt block + effort recommendation with rationale + bullet list of rewrites applied.
hmj1026/dhpk · ★ 2 · AI & Automation · score 71

Install: claude install-skill hmj1026/dhpk

# Prompt Optimize Rewrite a raw, informally-written task prompt into a model-aware, effort-calibrated version before you run it. ## When NOT to Use - Generic few-shot / chain-of-thought / prompt-template technique coaching with no model or effort selection involved — use `prompt-engineering-patterns`. - Auditing token/context-window budget or cache hit rate — use `dhpk-harness-budget`. - Actually executing the prompt — this skill only rewrites it; run the result yourself afterward. - Linting an existing `SKILL.md`'s own frontmatter/structure — use `dhpk-skill-health-audit` / `dhpk-skill-quality-judge`. ## Workflow 1. **Collect the raw prompt.** Take the user's pasted text verbatim, or `Read` the file if they pointed at one instead of pasting. Do not start rewriting yet. 2. **Detect and verify the target model.** Default to the current session's model, or use an explicit override from the user. Resolve the relevant library with `mcp__context7__resolve-library-id`, query it with `mcp__context7__query-docs`, then use `WebSearch` to locate the provider's official documentation and `WebFetch` to confirm the same parameter or capability there. These are explicit tools in this skill's `allowed-tools`; do not claim a live lookup was completed if one is unavailable. Follow the dated process in `references/live-model-verification-2026-08-05.md`; record the date and sources, or mark the recommendation `unverified` and keep it generic. 3. **Classify the task and pick an effort level