gh-issues

SKILL.md preview

---
name: gh-issues
description: >-
  GitHub issues (`gh issue` create/edit router). Single shipping entrypoint for issue create-or-edit. Gather requirements with an explicit Sharpen/Abort/Ship gate,
  dedupe against existing issues, ask user whether to update/create/abort, then execute through internal-write-gh-issue-commands.
  Optional multi-intent path: partition themes (including from `.cursor/plan.md` / `internal-read-plan-parts`), dedupe each theme, then one Proceed batch of ordered creates/edits.
---

**Orchestration:** [`internal-read-gh-issue-dedupe`](../../read/gh/issue-dedupe/SKILL.md) (`@gh-issues orchestration order` section).

# GitHub: issues (router)

Normative fences / full matrix: [`internal-read-gh-issue-list`](../../../read/gh/issue-list/SKILL.md), [`internal-write-gh-issue-commands`](../../../write/gh/issue-commands/SKILL.md).


**Public.** Top-level router for issue shipping. This skill owns create-vs-edit routing; there is no separate public create/edit path.

## Before batch (public, optional, sequential)

Run these only when they improve context for this invocation:

1. Use `@gh-issue-list` when you need a fresh inventory snapshot first.
2. Use `@gh-issue-view` when create-or-edit should be anchored to a specific issue context.
3. Preview only: list/dedupe hubs **`internal-read-gh-issue-list`** and **`internal-read-gh-issue-dedupe`** (Execution batch runs them in router order).

## Execution batch (internal, sequential)

Run this internal sequence in order (no runnable command fences in this public file). **Do not** run dedupe list/search until the user has chosen **Ship** on the pre-dedupe gate (or explicitly overrides).

1. `internal-read-gh-issue-spec`
2. `internal-read-gh-issue-preflight-qa` with **[`internal-write-plan-structured-qa`](../../write/plan/structured-qa/SKILL.md)** — iterate **Sharpen / Abort / Ship** until **Ship** or **Abort**
3. **Dedupe and description path (branch on intent count after clustering)** — see **`internal-read-gh-issue-dedupe`** for single- vs multi-candidate orchestration and **[`internal-read-plan-parts`](../../read/plan/parts/SKILL.md)** when themes come from a plan hub:
   - **Single-theme (default):** `internal-read-gh-issue-dedupe` → `internal-read-gh-issue-description` → `internal-read-gh-issue-labels` → `internal-read-gh-issue-projects-relationships` for **one** candidate.
   - **Multi-intent (optional):** follow **[Multi-intent clustering](#multi-intent-clustering-optional)**; then run `internal-read-gh-issue-description`, `internal-read-gh-issue-labels`, and `internal-read-gh-issue-projects-relationships` **per theme** that is not deferred, before one mutation summary.

## Multi-intent clustering (optional)

Enter only when useful—avoid extra steps for a clearly single-topic request.

**When to use:** after **Ship**, if **any** of: the user asks to split into multiple issues; the pasted request has **clearly disjoint** workstreams (unrelated bullets/sections); or a plan hub exists (`.cursor/plan.md` and/or linked `.cursor/plans/<slug>.md`) with **≥2** disjoint parts per **`internal-read-plan-parts`**.

**Steps:**

1. **Partition / cluster** — Prefer plan section boundaries when files exist; otherwise group bullets by shared root cause. Merge themes that are the same underlying work (issue #41: two bullets, one root cause → **one** issue).
2. **Inventory once** — `internal-read-gh-issue-list` (and checklist in **`internal-read-gh-issue-dedupe`**) over open issues; reuse that inventory for every theme. Optionally add targeted `gh search issues` per theme only when the open list is too large to scan fairly.
3. **Per-theme dedupe** — For each theme, working title + one-line intent vs inventory → `safe_to_create` | `likely_duplicate #n` | `ambiguous` (**never** skip dedupe for a theme that will **create**).
4. **Summarize** — Table: Theme id → overlap verdict → proposed action (**edit #n** / **create** / **defer**). Safe default: offer to **merge** near-duplicate

…