terryyin

dough-story-refinement

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. Continues a ready story into slice planning in the same workspace; `--refine-only` stops at the refined story. `--one-shot` refines a queued story without publishing an assignment and keeps the result for review, or lands it with `--auto-land`. Use for selected-story refinement, not broad problem decomposition, candidate selection, or slice sizing.

terryyin 4 Updated 2w ago

Resources

2
GitHub

Install

npx skillscat add terryyin/open-dough/src-skills-dough-story-refinement

Install via the SkillsCat registry.

SKILL.md

Story refinement

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.

A refinement invocation authorizes slice planning of its story and landing the
completed preparation under
preparation journey; it authorizes no
execution. Continue into planning 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 project's slice-planning workflow for the continuation, 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, then
continue a ready story into slice planning under
preparation journey unless an open
coordinator question or an explicit refine-only instruction (--refine-only)
stops it. When this session ends, follow
land at the end of preparation,
then close or retain the workspace. An execution instruction for this story
that arrives while its preparation is unlanded follows
hand off to execution in the same session
instead.

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, and where the draft is (its
workspace, and its result commit once committed). Do not recap what the seed
records. Then continue into slice planning under
preparation journey without a question or
approval request.

When the invocation ends at the refinement result, on an explicit refine-only
instruction (--refine-only) or under
one-shot refinement, the ready outcome
also gives 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, 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.