From Issue

/from-issue

Reverse-direction entry point. Most pipelines start from a blank story (impact-map); this skill starts from an issue that already exists in GitHub and back-fills the structured artifacts the pipeline needs.

When to use

• The team already opened an issue with prose acceptance criteria.
• You want to drive that existing issue through red-cycle / green-cycle without recreating it.
• The issue body has some structure (Goal / Acceptance / Files) but is not Gherkin.

If the issue body already contains Gherkin (Scenario: blocks), skip this skill and call /red <issue-number> directly.

Inputs

• <issue-number> — GitHub issue number in the current repo, or --repo owner/name <issue-number> to target another repo.

Workflow

1. Fetch the issue

gh issue view <issue-number> --json number,title,body,labels,milestone,assignees

2. Derive the slug

Build a us-slug from the title:

• If the title starts with a ticket prefix like T-235: or [FEAT-12], keep it: t-235-<rest-kebab>.
• Otherwise kebab-case the title and trim to 6 significant words.

Present the proposed slug to the user; accept an override.

3. Parse known sections

Look for these headings in the body. Use what you find; do NOT invent content.

If a checkbox is purely technical scaffolding ("oxlint clean", "snapshots regenerated", "i18n parity test green") classify it as non-business and surface it under ## Non-business gates — these become reviewer hints, not Gherkin rules.

4. Interview to fill the gaps

The pipeline needs five fields the issue rarely supplies; ask only for missing ones:

1. Actor — who triggers the behavior? Be specific (role, not "user").
2. Action — phrased as a verb against the actor's goal.
3. User-story sentence — As a <actor>, I want to <action>, so that <goal>.
4. Concrete data — pick representative values for any rule that mentions a threshold, list, or state. The pipeline rejects placeholders downstream; capture values now.
5. Test-level hint — likely @ui for frontend issues, @e2e for integration, @use-case for pure domain. Confirm.

Skip questions whose answer is unambiguous from the issue body.

5. Write context.md

Path: <paths.specs>/<us-slug>/context.md (from .atdd-pipeline.json).

Structure (same as impact-map, with two extra sections):

# <us-slug>

## Source
GitHub issue #<N>: <title>
URL: <html_url>

## Actor
<role>

## Goal
<copied from ## Goal section if present, otherwise filled from interview>

## Action
<from interview if not derivable>

## User story
As a <actor>, I want to <action>, so that <goal>.

## Business rules
- R-01: <text from acceptance checkbox 1>
- R-02: <text from acceptance checkbox 2>
- ...

## Non-business gates
- <technical checkbox 1, e.g. "oxlint clean">
- ...

## Implementation hints
- <files to create/modify, if listed>
- <architecture references, e.g. "ARCHI.md §6.5">

## Dependencies
- <blocked by T-234>

## References
- <PRD.md F-003>
- ...

## Milestone
<from issue.milestone if any, else "TBD">

## Default test level
<@use-case | @e2e | @ui from interview>

6. Tag the issue

Apply ATDD labels in-place (does not duplicate the issue):

gh issue edit <N> --add-label "atdd,us/<us-slug>"

If the user confirmed a test level, also add level/<use-case|e2e|ui>.

Do NOT remove the project's existing labels (scope:, track:, priority:, sprint:, etc.). They coexist with ATDD labels.

7. Seed issues.json

If <paths.specs>/<us-slug>/issues.json does not exist, create it with the imported issue as the parent AND the US (no separate parent / US split — the imported issue already plays both roles):

{
  "_imported_from": <N>,
  "parent": <N>,
  "us": <N>,
  "milestone": "<m>",
  "scenarios": {}
}

to-issues-atdd (re-)run later will populate scenarios after spec-generate. It will detect the _imported_from field and NOT create a new parent.

8. Handoff

Tell the user:

Imported issue # as <us-slug>. Next:

• /spec-generate to turn the business rules into Gherkin scenarios.
• Then /spec-review.
• Then /to-issues-atdd will create scenario sub-issues linked to #.

Idempotency

If <paths.specs>/<us-slug>/context.md already exists:

• Show a diff between current content and what would be written.
• Ask: overwrite, merge (add only new rules), or abort.
• Never silently overwrite.

Anti-patterns

• Do NOT promote technical checkboxes ("snapshots regenerated", "oxlint clean") to business rules. They are not behaviors the customer cares about; they are CI gates. They land under ## Non-business gates.
• Do NOT close or re-title the imported issue. The team's history stays intact.
• Do NOT split an issue into multiple US during import. If the issue covers two unrelated stories, ask the user to file a second issue first.