AGENT SKILL

highway-objectives

Manages the repository-wide business objective baseline and its stable traceability references. Use when highway-setup delegates the objective baseline, after Profile finishes or when setup runs again.

License 许可
Apache-2.0
最近更新
2026-11-15
Stars
★ 0
Forks
0

01 前言 Front Matter

name
highway-objectives
license
Apache-2.0
metadata
{ "usage": "Run highway-objectives.", "version": "8.8.0" }
description
Manages the repository-wide business objective baseline and its stable traceability references. Use when highway-setup delegates the objective baseline, after Profile finishes or when setup runs again.

02 资源文件 4

03 SKILL.md 正文

highway-objectives

Purpose

Collects business objectives when setup delegates this skill.

When to use

  • highway-setup delegates this skill after Profile returns Finished.
  • highway-setup delegates this skill when setup runs again.

When not to use

Do not create Controls, NFRs, Capabilities, or governance rules. Do not treat a recommendation or a queued title as a captured Objective. Do not report Status, Summary, Next Action, Blocking Reason, Missing, Complete, or Blocked. Do not emit Setup's outcome transition. Setup already emitted it.

Inputs

  • library/knowledge/profile.md when present.
  • library/governance/objectives.md when present.
  • library/objectives/OBJ*.md when present.
  • assets/objective-record.md.
  • assets/objective-catalog.md.
  • The highway-experience skill, activated first.
  • Delegation from highway-setup.

Outputs

The complete structure is assets/objective-record.md and assets/objective-catalog.md.

  • Updated Objective records and catalog after a review edit.
  • New Objective records and catalog after the guided flow writes them.
  • Finished returned to highway-setup after review acceptance or a successful guided write.
  • Finished is not shown to the person.
  • One closing line after review acceptance or a successful guided write, when the person ran this skill directly.

Procedure

Activate the highway-experience skill before any output the person sees, and follow it for Highway's identity, the Experience Standard, and the storage contract.

User-visible interaction follows the Highway Experience Standard. Repository ownership stays with the person.

Emit only the active template. Follow this order: review when the catalog exists; otherwise recommend, then queue or ask, then capture, then identifier, title when needed, and write. Both paths end at close. Follow the Highway Experience Standard for attribution and acceptance wording. Use the accepted Organization Name from the Profile. Use Repository Name when no Organization Name is accepted. Do not ask for Profile facts already accepted.

Before a step in this table, read its file and follow it to the end.

File Steps
references/proposal.md recommend, more, queue, ask, capture
references/record.md identifier, title, write

Entry

  1. Read library/governance/objectives.md. Then read each record it indexes at library/objectives/[ID].md, with "Read a file".
  2. If the catalog and every indexed record are present and readable, go to review.
  3. If the catalog is absent, go to recommend when profile.md is readable. Otherwise go to ask.
  4. If the catalog or an indexed record is unreadable, or a catalog row has no record, stop. Say the Objectives could not be read. Leave every file unchanged. Do not return Finished.

Classify

Classify the latest reply before any other action.

  • select-all: affirms every recommendation in the current set and adds no separate outcome.
  • select: names one or more recommendations in the current set, by title or by number.
  • own: supplies an outcome that is not a displayed title.
  • more: asks for more suggestions.
  • review-now: asks to review the queue, or affirms a queue prompt that asks whether to review.
  • edit: adds, removes, corrects, splits, combines, or replaces substance.
  • accept: affirms the current capture or review body and adds no fact.
  • unknown: the person says they do not know.
  • question: the person asks for an explanation.
  • decline: the person explicitly drops the active set.

On a recommendation set, "looks good" is select-all. On a queue prompt that asks whether to review, "looks good" and "let's review" are review-now. On a capture or on review, "looks good" is accept.

review

Show every Objective in catalog order. Omit frontmatter and omit the catalog. For each Objective emit:

[Title]: [Statement]

Success looks like:

  • [Success measure]

Why it matters: [Rationale]


Put --- between Objectives. Do not put --- after the last Objective.

When the catalog has no records, emit: No objectives are recorded yet.

Then emit the validation block.

  • edit: emit one consequence sentence. Apply the edit to the named Objective. Preserve every Objective the person left unchanged. An added outcome becomes a new record. A removed outcome deletes that record file and keeps its catalog row with status removed, so the id stays taken. Derive the id before writing. When the id is taken for a different title, go to title and do not write yet. Otherwise rewrite the affected records and the catalog. Show the updated Objectives and emit the validation block. Stay on review.
  • accept: go to close.
  • unknown: show the Objectives and emit the validation block. Stay on review.
  • question: answer in one sentence. Show the Objectives and emit the validation block. Stay on review.

Validation block:

What would you add, correct, or remove?

If this is accurate, just say so.

close

When highway-setup delegated this skill, return Finished. Emit nothing else.

Otherwise emit one line and stop. When at least one Objective is active, emit:

Your Objectives are recorded.

When no Objective is active, emit:

No Objectives are recorded yet. Run highway-objectives when you have one to add.

Verification

  • An existing readable catalog is shown with What would you add, correct, or remove?
  • Acceptance of that review goes to close.
  • Run by highway-setup, close returns Finished and emits nothing else. Run directly, it emits one closing line and no question.
  • A review edit rewrites the affected record and catalog, then shows the updated Objectives with the same question.
  • An absent catalog and a readable Profile begin with a recommendation set.
  • Records and the catalog are written once, after the set is accepted or declined.
  • Written records match assets/objective-record.md.
  • The written catalog matches assets/objective-catalog.md.
  • Setup receives Finished and the person does not see that word.

Error Handling

  • An unreadable catalog or record stops the skill, leaves every file unchanged, and does not return Finished.
  • A failed review rewrite leaves every file unchanged, says what failed in terms the person can act on, and stays on review.
  • A failed guided write leaves the baseline unchanged, says what failed in terms the person can act on, and stays on write.
  • An absent or unreadable Profile uses ask and does not invent organizational facts.

Example

/highway-setup

04 同仓库更多 Skills