AGENT SKILL

advisor

Consult the fable-advisor agent for a design/plan review. Use when the user types /advisor, says アドバイザーに聞いて / Fableに確認 / セカンドオピニオン, or proactively BEFORE implementing a large plan (new tables/RLS, board algorithm rewrites, concurrency-sensitive changes, cross-cutting refactors) instead of asking the user to review the plan.

License 许可
MIT
所属仓库
l4l4dev/Storylane
最近更新
2026-09-21
Stars
★ 0
Forks
0

01 前言 Front Matter

name
advisor
description
Consult the fable-advisor agent for a design/plan review. Use when the user types /advisor, says アドバイザーに聞いて / Fableに確認 / セカンドオピニオン, or proactively BEFORE implementing a large plan (new tables/RLS, board algorithm rewrites, concurrency-sensitive changes, cross-cutting refactors) instead of asking the user to review the plan.

03 SKILL.md 正文

Advisor consultation

Spawn the fable-advisor agent (Agent tool, subagent_type: fable-advisor, synchronous) with one focused question.

Package the prompt as:

  1. The question — one decision to make ("この計画で進めてよいか", "A案とB案 どちらか"), not "review everything".
  2. Context — current Backlog task id, your plan or a diff summary, the files you intend to touch.
  3. What you already checked — spec sections read, tests run, related tasks looked at.

Rules:

  • Keep the prompt under ~40 lines. The advisor reads the repo itself — do not paste whole files.
  • Fable time is metered on the user's plan: batch related sub-questions into one consultation instead of spawning repeatedly. Do not use the advisor for questions the project instructions or the spec already answer.
  • Relay the advisor's verdict to the user (verbatim conclusion + key corrections) before acting on it.
  • If the verdict changes a spec or Backlog task, you make those edits per its instructions — the advisor never writes.
  • If the fable model is unavailable (plan window closed), rerun the same Agent call with model: opus and tell the user the advisor ran on Opus.

04 同仓库更多 Skills