| name | enhance-prompt |
| targets | ["*"] |
| description | Normalize any raw request into a grounded, structured brief before acting. The auto pre-step for skills and subagents: extract the real intent, gather scoped context (issue tracker + roadmap + docs/ADRs + catalog), surface only blocking ambiguities, and emit objective / scope / constraints / deliverables / guardrails. Routes build-type asks into `enhance-intent`. Skips asks that are already precise. Use on any fuzzy, broad, or multi-part request, or at the start of a skill or agent that received a raw ask.
|
enhance-prompt
Turn a raw, rambling ask into a brief a stranger could execute - before any file is touched. A one-line ask is a starting point, not a spec; a wrong reading executed confidently wastes far more than the enhancement costs.
Runs as a pre-step: skills and the orchestrator apply it to the incoming ask, then act on the enhanced brief - not the raw text.
Run it, or skip it
- Run when the ask is fuzzy, broad, multi-part, or missing scope / target / success criteria.
- Skip - act directly - when the ask is already precise and self-contained: a specific edit, a single factual question, a mechanical change with an obvious target. Enhancing a clear one-liner just burns tokens.
Match effort to the ask: a small task gets a one-line restatement, not a full brief.
Method
- Restate the intent in one line. If your restatement might be wrong, gather context or ask - don't guess.
- Classify: build (feature / adapter / page / route) · review or audit · debug · refactor · docs · research · ops. For a build ask, hand off to the
enhance-intent MCP tool (server oss) - it adds the platform catalog, requirements checklist, and the consumer pnpm gen playbook - and stop here.
- Gather scoped context - targeted, never a data dump (token budget matters). Pull only what changes the plan: the issue tracker / roadmap,
docs/ + docs/adr/, the catalog MCP tools (catalog-overview, list-adapters, describe-module), and the code itself. Scope first, then fetch.
- Surface blocking ambiguities only - the questions whose answers change what you build. Resolve the rest with stated defaults. Don't interrogate.
- Emit the brief, sized to the task.
The brief
- Objective - the outcome, in priority order.
- Scope - what's in; what's explicitly out.
- Context - the load-bearing facts found, with links (issue, doc,
file:line).
- Constraints - conventions, clean-architecture layering, minimal-dependency + token budget, the public / open-source surface.
- Deliverables - what "done" produces.
- Guardrails - what not to touch; reversible-only; confirm-before-X.
- Open questions - only the blocking ones.
Modes
- Return for approval (default): for non-trivial or ambiguous asks, present the brief, get a nod, then execute. Silently running a wrong reading multiplies the waste.
- Enhance then proceed (internal dispatch): the orchestrator enhances once at the top and passes the crisp brief to the subagent. Subagents act on the brief; they do not re-enhance their own input.
Rules
- Intent-preserving: clarify the ask, never invent scope the user didn't want.
- Proportional: the brief is as short as the task allows.
- One pass at the top: enhance before context-gathering and before delegation, so the gather is targeted and the agents get a clean brief.