What problem does it solve? Half-formed ideas like "should we adopt X" or "how would we build Y" often get answered from memory, turn into endless option surveys, or lose their decisions in chat history. This Skill turns a vague idea into a grounded, decision-complete proposal by researching the web and codebase, separating established facts from human decisions, and converging on a saved prompt artifact ready for the planning layer. ## Core Features & Use Cases - Tiered right-sizing: A triage gate classifies ideas as Tier 0 (answer directly), Tier 1 (lightweight single-round proposal), or Tier 2 (full multi-round loop), so small ideas never get bloated documents. - Parallel grounded research: Fans out web research (WebSearch/WebFetch or the optional deep-research skill) and codebase exploration agents concurrently, quantifying real surfaces instead of estimating. - Human decision cadence: Uses recommendation-first AskUserQuestion gates to confirm the objective, choose among candidate approaches, and resolve open decisions, recording each answer in the artifact. - Living artifact with planning handoff: Authors a date-free proposal-<slug>.md saved prompt via the plan-agent:prompt skill, then hands off to /plan-agent:implementation-plan with --from-prompt. - Use Case: A developer asks "should we align our markdown components with Google's DESIGN.md spec?" The skill fetches the real spec, maps the codebase's token layer, presents candidate approaches, locks decisions, and produces a proposal with workstreams, risks, and a sized roadmap. ## Quick Start Ask the agent to turn your idea into a proposal, for example: run /plan-agent:build-proposal with a description of the idea you want evaluated.
