issue

name	issue
profiles	["kanban"]
description	Turn an issue into finished work. Use when a GitHub issue URL (e.g. github.com/owner/repo/issues/123) is pasted, or when the pasted text is clearly the body of an issue (title + description, "Steps to reproduce", "Expected/Actual behavior", labels). Also fires on "/issue", "make a task from this issue", "implement this issue", "do this issue". Converts the issue into one or more kanban tasks and then drives them to done.
license	MIT OR Apache-2.0
compatibility	Requires the `kanban` MCP tool to persist tasks and delegates to the `task`, `plan`, and `finish` skills. Fetching an issue from a bare URL uses the `gh` CLI (`gh issue view`) when available, falling back to `WebFetch`; pasted issue content needs neither. Will not function on a harness that does not expose `kanban`.
metadata	{"author":"swissarmyhammer","version":"{{version}}"}

Issue

Convert an issue — pasted as a link or as its content — into kanban task(s), then drive them to done.

$ARGUMENTS

Orchestrator only — does not research, write tasks, or write code itself. It normalizes the issue into text, then delegates: /task or /plan to create the card(s), /finish to complete them.

1. Recognize the input

Treat the input as an issue when it is either:

• A GitHub issue URL — matches github.com/<owner>/<repo>/issues/<number> (also accept gh-style owner/repo#123).
• Issue-shaped text — a title plus a body, especially with sections like Steps to reproduce, Expected behavior, Actual behavior, Description, a checklist, or label/severity lines. The user pasting a chunk of bug-report prose counts.

If it is genuinely ambiguous whether the paste is an issue or a free-form request, prefer this skill when it reads like a report of work to be done. If it is clearly a question or a discussion, this skill does not apply.

2. Normalize to issue text

• URL → fetch the issue. Prefer the authenticated, structured gh CLI:

  gh issue view <url-or-owner/repo#number> --json title,body,labels,comments
  

  If gh is unavailable or unauthenticated, fall back to WebFetch on the URL. If both fail, STOP and tell the user — do not invent the issue contents.

• Pasted content → use it directly; no fetch needed.

Read the comments/discussion when present — they often carry the real acceptance criteria. The normalized result is the issue's title + body + relevant discussion as plain text, used verbatim as the basis for the task(s).

No GitHub coupling. Do not store the issue URL on the card and do not write anything back to GitHub (no comments, no labels, no close). The card is created from the issue's content; once created it is an ordinary kanban task.

3. Size, then route to /task or /plan

Judge the normalized issue against the task sizing limits (one concern, 2–4 files, ≤5 subtasks):

• Single concern, fits one card → /task <issue text>. The task skill researches the codebase and writes one well-formed card (What / Acceptance Criteria / Tests). Capture the resulting short_id.
• Multi-concern or too large for one card (multiple independent changes, >5 natural subtasks, spans many files) → /plan <issue text>. The plan skill decomposes it into several right-sized cards linked with depends_on. Capture the resulting short_ids.

Do not hand-write task descriptions here — that is /task and /plan's job, including the architecture research and the Task Standards template.

4. Finish

Hand the created id(s) to /finish, which loops implement → test → commit → review until each lands in done:

• One card → /finish <short_id> (single-task mode).
• Several cards from /plan → /finish over the batch (e.g. the project/tag the plan created), so every card is driven to done in dependency order.

Report back: which issue was ingested, the card(s) created (by short_id), and the final /finish outcome (done, or any task reported stuck).

Constraints

• Kanban is the single source of truth — no TodoWrite/TaskCreate. Cards come only from /task or /plan.
• Reuse, don't reimplement. This skill adds recognition + normalization + chaining; all research, task-writing, and the implement loop live in the delegated skills.
• One issue per invocation. If several issues are pasted at once, take the most important and tell the user to re-run for the rest.

{% include "_partials/short-ids" %}