Story refinement

SkillDev tools

Lets your agent clarify selected user stories by recording their goal, scope, and key examples before planning work.

Use Story refinement in Claude, ChatGPT or Ahel Desktop

Free. Sign in, add Story refinement and connect your AI. About a minute.

Also: Claude Code · Cursor · Codex

Then ask your AI: use the Story refinement skill

Details

Instructions available. Your AI can read the instructions. Execution depends on the setup they require.

Add Ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

Story refinementStart free
About this skill

Clarifies selected stories before execution planning by establishing goal, scope, and key examples in each story's seed, distinguishing promises from rejection constraints. Adds UI or architectural detail only when needed. Use for selected-story refinement, not broad problem decomposition, candidate

What this skill tells your AI

The instructions your AI receives, as published by terryyin/lizard in .agents/skills/dough-story-refinement/SKILL.md and read by Ahel’s review.

Build shared understanding of one selected story, or several related stories whose boundaries need discussion. Record Goal, Scope, and Key examples in each story's seed.

Choose the workflow

If the parent problem, candidate selection, or story ordering needs reconsideration, use dough-story-decomposition. For smaller or clearer slices, use the project's execution-plan refinement workflow on the existing plan.

Refinement alone does not authorize planning or implementation. When the user explicitly requests execution planning, hand off one understood story to the project's planning workflow and continue without repeating answered questions.

Resolve required context

Identify this project's root, selected story links and seeds, and relevant prior decisions. When a seed is missing, resolve the canonical seed directory, ID and filename conventions, required metadata, and stable story-anchor convention. Resolve project paths from that repository, not this skill's location.

Resolve the execution-planning workflow only for a requested handoff, and ADR context only when the architectural concern requires it under the reference below. If required context or a linked dependency is unavailable, name what is missing and stop the affected activity. Do not invent project paths or decisions.

Refine and report

Read and follow planning scope and lifecycle for the conversation, scope decisions, optional UI and architecture detail, seed updates, and cleanup after implementation. When your instruction carries an established preparation, follow established preparation instead of the workspace and announcement steps below. When the request explicitly selects one-shot (--one-shot) for a queued story, follow one-shot refinement instead of the announcement step and the default disposition. Before writing to a story's seed, establish or reuse the required workspace under preparation workspace, then, for a queued story, announce the preparation assignment; refinement discussion and clarifying questions need neither on their own. After the seed or correction-home write records goal, scope, and key examples for a work item with a known identity, apply record preparation facts.

When the request includes options such as --explore, read refinement options and apply the selected options' instructions within this workflow. Options in the same group, an entry of that file's groups list, are exclusive; if a request names more than one of them, stop and report the conflict, naming the group's label and the flags the request named from it. Without options, refine straightforwardly.

Report each selected story under report the refinement outcome below. Apply the preparation disposition keep or discard decision, then close or retain the workspace, when this session ends.

Report the refinement outcome

End each selected story with exactly one outcome. When several stories are refined together, judge each independently: a clear story keeps its result, and another story's open decision stays with that story.

  • Ready for slice planning — goal, scope, and key examples are recorded, no decision you need from a person remains, and the work needs planned execution.
  • Flawless — ready for execution — the same, and the story fits one planless slice under slice sizing, with no decisive premise, as slice planning defines it, left unobserved and no probe needed.
  • Needs human engagement — at least one response from a person is required.

A ready outcome states the outcome, the story link, where the draft is (its workspace, and its result commit once committed), and one concrete next step: slice planning in that workspace, or, for Flawless, execution with an explicit instruction to skip slice planning. When the story already has an associated plan, the next step uses that plan instead of creating another. Name the pending draft and any Preparing assignment as information, not as a request to keep them. Do not recap what the seed records, and end without a question or approval request.

Needs human engagement lists each expected response separately: what must be answered or decided, who decides, your recommended answer when you have one, and what continues once it is given. It covers an open goal, scope, or constraint decision; a boundary change such as splitting, merging, or dropping the story; missing required context; and a stopped write, recording, or landing. Claim no readiness. When required context is missing before the seed write, name the missing input and who can supply it, say which activity resumes afterwards, and do not claim that refinement was recorded.

The outcome is reported, not recorded. Recording still follows record preparation facts: refined, keeping an existing selected approach and plan association, otherwise an unselected approach. Flawless grants no planless authority; the developer supplies it by running the skip-planning next step. Reporting an outcome neither creates nor renews ready and never selects planless; assess readiness at preparation completion still owns the assessment of the current story and any plan.

Signals

GitHub stars
3k
Forks
297
Last commit
Sep 2026
Advanced
Item type
skill
Key
dough-story-refinement-terryyin
Source
github.com/terryyin/lizard