hora-spec-horizon
Stage 2 of /hora-spec. Narrow the release to the fewest use cases somebody needs, and split the rest into deferred-with-a-seam and never-built. Writes the implementation scope, the build order, and the version's own acceptance criteria. Invoked by /hora-spec, or directly.
openreachtech · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add openreachtech/hora-core --skill hora-spec-horizon技能说明
name: hora-spec-horizon description: Stage 2 of /hora-spec. Narrow the release to the fewest use cases somebody needs, and split the rest into deferred-with-a-seam and never-built. Writes the implementation scope, the build order, and the version's own acceptance criteria. Invoked by /hora-spec, or directly.
hora-spec-horizon
Stage 2 of /hora-spec. Decide what this release builds, what it deliberately leaves for later, and what it will never build — and keep the three apart.
Read ../hora/references/structure.md and ../hora-spec/references/principles.md first. ../hora/references/asking.md fixes how anything here is put to a person — a check, a proposal or a question, each with the question tool as its default. ../hora-spec/references/stages.md is the authority on this stage's exit condition.
This stage reads nothing new, and one thing stage 1 wrote down for it. What already exists is stage 1's finding; what to build next is a decision, and no repository holds one. Every question here is a question or a proposal — there is nothing to put up as a check. The one thing to read is .hora/spec/<version>/_stages.md, where stage 1 recorded the criteria that reached past the feature they were drafted for ("The version's own acceptance criteria", below).
What this stage decides
which of stage 1's use cases this release serves
which are deferred, and what unblocks each one
which are never built
which seams the deferred ones need kept replaceable
in what order the release builds what it builds
which behaviors belong to the version rather than to any one feature
What it must not decide
| Whose it is | |
|---|---|
| a new use case | stage 1. If one turns up here, go back and state it there first |
| how a seam is built | stage 4, and the package's own skills |
| whether a deferred item is technically possible | nobody, yet. It is deferred |
| the version number a deferred item lands in, where nobody has decided one | leave it as "no version yet", with the trigger |
The narrowing
A release carrying too much is the default outcome, not a risk (../hora-spec/references/principles.md). Everyone wants everything first, and nothing about writing a document resists it. So this stage resists it.
The question that does the work is not "is this important?" — everything is important, and asking it produces a list identical to the one you started with.
which use case is impossible without this?
| The answer | What it means |
|---|---|
| a use case from stage 1, named | it is in. That use case is the reason |
| "it would be nice to have" | it is deferred. Ask what would make it necessary |
| a use case nobody stated | go back to stage 1. Either the use case is real and belongs there, or the feature has no reason |
A section carrying <!-- baseline: inventoried --> states no use case by declaration, and is not sent back to stage 1 for it (../hora/references/spec-format.md, "baseline"). Its silence is somebody's decision, not a hole.
Then say the number out loud. "This release has eleven features and four of them serve one use case each" is a sentence nobody says to themselves. Propose the split — which four go now, which seven follow, and which use case each group completes.
If the answer is still "all of it", record it and carry on. State it once, propose the narrowing, and if the decision stands, write a scope question naming what was proposed and who declined it. The decision belongs to whoever asked for the product; saying nothing does not.
A narrowing that was declined names when it gets read again
"Never nothing" is this stage's rule about deferred items. It applies to this
stage's own decisions too. A scope question records what was proposed and
who declined it, and that is where the record stops today: nothing states when
the decision would be worth reading again, and nothing reads it.
The condition does not go in the question. A condition is a decision, and
.hora/ holds no decisions (../hora/references/structure.md, "Where a lever
lives"). It goes into the implementation scope, in the owner's own words, shown
and approved like every other line of that section
(../hora/references/spec-format.md, "4. Implementation scope"):
### Built this time (1.0.0)
- ... the features ...
Reconsider 1.0.0's scope when: milestone 1 closes with more than four features
unaccepted, or any one feature is sent back twice.
Three properties make it a condition rather than a wish. It names the version
it binds — the reach a lever states where it is declared
(../hora/references/structure.md, "Where a lever lives"): this version alone,
dead the moment a later scope section is written. It is measurable against
what .hora/ already records — the plan's boxes read with §14's milestones
("milestone 1 closes" is every feature §14 places there standing [x]), the
blocks in .hora/acceptance/, the question file. And somebody reads it:
/hora-plan walks the line on re-entry and raises it once
(../hora-plan/SKILL.md, "6. Reconcile on re-entry"). "If 1.0.0 overruns" has
none of the three.
Propose it; do not write it unasked. Where the owner names no condition, record that too — a declined narrowing with no re-reading is a decision somebody made with their eyes open.
The three lists are three lists
Nothing about this is a formality. The two kinds of out-of-scope produce opposite designs, and /hora-plan carries each into every feature file as a constraint:
| The list | What the design does | What a feature file gets |
|---|---|---|
| built this time | build it | the feature itself |
| out of scope for now (to be built later) | leave a seam. Keep what is behind it replaceable | Constraint: leave room for … |
| permanently out of scope | do not abstract it. Exclude it | `Constraint: … is permanently out of sco |