Improve Prompt

Enrich the user's draft prompt into a complete, ready-to-use prompt by applying the prompt rule pack. Enrichment is not generation from nothing: strengthen the draft the user already wrote, fill structural gaps, tighten ambiguity — never redirect the stated intent.

Pipeline

Run these four steps in order.

1. Load the rule pack

Read ${CLAUDE_PLUGIN_ROOT}/rules/prompt-rules.json (equivalently ../../rules/prompt-rules.json from this skill's base directory). Every enrichment decision traces to a rule in that pack; the pack maps each rule to the research finding(s) behind it via finding_refs, resolvable in ${CLAUDE_PLUGIN_ROOT}/rules/findings-index.json.

2. Confirm the artifact kind

The invocation of this skill is the explicit classification: the draft is a prompt. If the draft clearly reads as a different artifact kind instead — a goal (a desired end state) or a loop (a recurring cadence or open-ended monitoring task) — surface that mismatch as the first clarifying question. Never silently convert it. In default single-turn mode, honor the explicit invocation: enrich the draft as a prompt anyway, with the mismatch question folded in as the first line of the returned artifact. In --interactive mode, ask the mismatch question first, before any other gap.

3. Detect gaps

Run the draft against every rule's gap_check. Produce an internal gap list of {rule_id, finding_refs, description} records, carrying every finding ref the rule cites. This one list feeds both elicitation modes below — never maintain mode-specific notions of what counts as a gap.

4. Elicit and synthesize

Default (single-turn): fold each gap into the returned prompt as a concise clarifying question at the point where the answer belongs. Nothing is invented; nothing is a placeholder token.

--interactive (only when the argument includes the flag): render the same gap list as AskUserQuestion calls — at most 4 questions per call, batched until every gap is resolved — then synthesize the enriched prompt with the answers folded in as real content. If the user declines or skips a question, fall back to folding that gap into the output as a clarifying question — never invent an answer.

Synthesize by applying every applicable rule from the pack. Structured output is the default: emit the prompt with explicit XML-tagged sections (role / context / task / constraints / output-format — the section set that targets the prompt's purpose). When the prompt's output will be machine-parsed, include an explicit JSON Schema output contract instead of prose format descriptions. Freeform prose output is the exception and requires a genuine reason — a judgment made internally, never stated in the output.

Output format

Return ONLY the improved prompt inside a code block. Nothing else. Exception: when the user explicitly asks for the rationale or grounding behind changes, append the per-change citations (rule id → finding @id) after the code block.

[improved prompt here]

Example

Improved:

<role>You are a customer-insights analyst.</role>
<context>Feedback arrives as free-text survey responses. Which product or time period this covers is unstated — see the question in <task>.</context>
<task>Summarize the recurring themes in the provided feedback. Clarify first: which product and time period should this cover, and roughly how many responses are in scope?</task>
<constraints>Base every theme on at least two distinct responses; quote one representative response per theme; do not speculate beyond the provided text.</constraints>
<output-format>A ranked list of themes, each with a one-sentence summary, a count of supporting responses, and one representative quote.</output-format>

Critical rules

• Return ONLY the code block with the improved prompt — no analysis, no explanations, no "Changes Made".
• The improved prompt must be COMPLETE and READY TO USE immediately, including any clarifying questions needed to proceed.
• NEVER use placeholders like [LANGUAGE], [FILE], [TOPIC]; NEVER create templates. If information is missing, fold a concise clarifying question into the prompt; do not invent details.
• Preserve the original intent: add specificity, structure, and missing elements; never substitute a different task.
• Keep the improved prompt concise but complete.
• When the user asks why a change was made ("why did you add X?"), cite the specific rule id and its finding_refs — resolved to the full finding @id and source URL via rules/findings-index.json — not a restated heuristic.