/plan-feature-from-issue

name: plan-feature-from-issue
user-invocable: false
version: 2.0.0
author: "Gabriel Trabanco <gtrabanco@users.noreply.github.com>"
license: MIT
description: >
  Internal step of plan-feature: turn a feature-request issue into a scoped,
  sized, roadmap-mapped SPEC **product half** (capability closure satisfied)
  with Closes #N traceability.

Plan Feature — From Issue (internal)

Convert a feature-request issue into the project's planning artifacts, keeping a clean issue → SPEC → PR(Closes #n) trace. Writes the SPEC's **product half** (same two-halves convention `design-feature` uses) and must satisfy capability closure before handing off — a thin issue does not get a shortcut around it.

**This skill stops at the Product half.** It designs, then the unit goes to `review-spec` for an independent Product review; engineering planning is a different authority's turn. Composing `plan-feature-scaffold` in the same breath as the design it just wrote is the bypass this separation exists to close — the author of a Product half cannot be the one who decides it is ready to build on.

When to use

• The `plan-feature` router calls this when the input is a GitHub issue (or

`--from-issue N`) that describes new product capability.

If the issue is a **bug or tech-debt**, stop and route it: `triage-issue` to classify, then `plan-fix` + `execute-phase --fix`. This skill is for genuine features only.

Step 0 — Discover the project (always first)

Per the agent guide's **Workflow conventions** + **documentation map**, then read what THIS skill needs: the feature SPEC template, the roadmap, and the issue/PR templates (`.github/ISSUE_TEMPLATE/`, `.github/PULL_REQUEST_TEMPLATE.md`) so the SPEC mirrors the fields reviewers expect. Then read the issue (forge CLI per the project's Workflow conventions — examples use `gh`):

gh issue view <N> --json number,title,body,labels,state,comments

Process

1. **Classify first.** Confirm it is a feature. Not a feature if it describes a defect, regression, duplicated code, perf debt, or carries a "when to fix / trigger" clause → hand to `triage-issue`. State the verdict explicitly. 2. **Normalize language.** If not in the project's docs language (this repo: **English**), translate before drafting any artifact. 3. **Map to the roadmap.** Assign the next number + slug. Identify dependencies and conflicts with existing features, coupling/migration risks, and whether it should instead extend an existing feature. 4. **Close product-half gaps proactively.** Compare the issue against what a complete SPEC **product half** needs (goals, scope in/out, business goals, i18n/SEO/a11y/pricing per the docs map, a UI design reference when the feature has a UI surface), probing the same fixed **vagueness rubric** `design-feature`'s interview uses: affected users/roles · error & edge states · data shape · boundaries & limits · out of scope · success criteria — each slot filled or explicit `n/a: <reason>`. For each genuine gap you can't safely default, ask the user **one question per turn, never batched**, each with a recommended default; never ask what the issue or docs already answer. **Structural hand-off threshold:** if ≥ 3 rubric slots remain unfillable from the issue plus the answers so far, stop and hand the feature to `design-feature` (the thin-issue rule below, now with a fixed trigger) instead of continuing to interview here. 5. **Satisfy capability closure.** Walk the same fixed checklist `design-feature` uses (per entity: CRUD + state transitions, each with UI + API + test, or explicit `n/a: <reason>`; per capability: entry point + ACL; per role: assigned/revoked/viewed where) into the SPEC's `## Capability closure` and `## Acceptance criteria`. **A thin issue that doesn't carry enough to fill it is not a shortcut around the gate** — hand it to `design-feature` (compose in-turn only at ≥ this skill's tier, per *Guardrails*; otherwise hand off with `run /design-feature <slug>` and stop here) rather than stamping `designed` on a hollow closure. 6. **Size it.** Estimate `XS / S / M / L` (scale defined in the SPEC template) and record it in the SPEC. XS/S → the SPEC is the only planning artifact (single-pass execution); M/L → full artifact set. If L, propose splitting. 7. **Produce the SPEC product half.** Fill it and stamp `## Design status: designed` once closure is complete; set the roadmap row (added at `idea` first if it didn't exist) to `defined` in the same edit — the same `idea → defined` transition `design-feature` owns, performed here when this skill is the one that satisfies closure. Then run the `stage: spec` readiness preflight from the internal [`evidence-grounding`](<../evidence-grounding/SKILL.md>) capability, mint the current `artifactRevisionId`, and **stop**: this skill never continues into the engineering half, never promotes the row past `defined`, and never composes `plan-feature-scaffold` in this turn. The `plan-feature` router may scaffold only after `review-spec` returns a current `spec-review-pass` receipt bound to these exact bytes. 8. **Wire traceability.** Record `#N` in the SPEC; the PR body must include `Closes #N` so the issue closes on merge. 9. **Hand off — return exactly** (fixed completion report, back to the router):

   ISSUE #<N> → SPEC <slug> — size: <XS|S|M|L>
   Verdict: feature (not bug/debt — else this would have routed to triage-issue)
   Gaps closed: <n> asked / <n> defaulted (logged)   Closure: designed | handed to design-feature
   Readiness: READY-FOR-REVIEW | NEEDS-EVIDENCE | NEEDS-DESIGN   Artifact revision: <id>
   Traceability: Closes #<N> wired
   → review-spec next (engineering planning is gated on its receipt; do not scaffold here)

Guardrails

• Don't silently expand scope beyond the issue — surface additions as proposals.
• Don't open the feature branch or write c