create-ticket

name	create-ticket
model	haiku
description	Interactively create GitHub issues (Bug, Feature, Task) for the current repository. Infrastructure changes are Tasks auto-labeled `ops-request`. Parses optional args for type and description, asks targeted questions, assigns labels, and creates the issue via gh CLI.
alwaysApply	false
metadata	{"author":"project","version":"1.0"}

Create a GitHub issue in this repository. Guides the user through an interactive flow to gather all required information, then creates the issue via gh issue create.

Prerequisites

Before starting the flow, verify gh CLI is available and authenticated:

gh auth status

• If gh is not installed → stop and tell the user: "GitHub CLI (gh) is required. Install it: https://cli.github.com/"
• If not authenticated → stop and tell the user: "Run gh auth login to authenticate first."
• If authenticated → proceed to Step 0

Usage: /create-ticket [type: description]

Examples:

• /create-ticket — fully interactive
• /create-ticket bug: login page crashes on Safari
• /create-ticket feature: add bulk delete for models
• /create-ticket task: refactor Sidebar for reuse in entity lists
• /create-ticket infra: add LOG_LEVEL to prod

---

Step 0 — Parse Args

If the user provided arguments after /create-ticket, parse them:

• Look for a type prefix: bug:, feature:, task:, infra: (case-insensitive)
  • bug: → Bug
  • feature: → Feature
  • task: → Task (general engineering)
  • infra: → Task + force ops-request label (shortcut for infra change)
  • If found → set issue type, use the rest as a seed for the title/description
  • If not found → treat the entire arg string as a description seed and ask for type
• If no args at all → start fully interactive from Step 1

When args provide a description seed, use it to:

• Generate a proposed title — concise, under 80 characters
• Generate a proposed summary — a short paragraph followed by key points as a bullet list. For bugs, also draft the actual/expected result from context.
• Pre-fill relevant body fields where obvious
• Auto-assign labels ONLY when the description makes them 100% clear
• For ANY label where assignment is not certain from the description, ask the user directly

---

Step 1 — Issue Type

Skip if parsed from args.

Use AskUserQuestion to ask:

"What type of issue do you want to create?"

Options (these match the repo's GitHub issue types):

• Bug — Report a problem or defect
• Feature — Request a new feature or enhancement
• Task — Engineering work that is not a user-facing feature (refactor, reuse, tech debt, cleanup, test/build improvements, AND infrastructure changes — env var, secret, config, deployment setting)

Infrastructure changes are NOT a separate top-level type. They're a Task that gets auto-labeled ops-request. Detection happens in Step 3 from context keywords (env var, secret, prod/uat, deployment, config, LOG_LEVEL, etc.).

---

Step 2 — Title & Summary Proposal

2a. Optional Code Research

Before drafting the proposal, check whether the user's input references code concepts that would benefit from codebase investigation.

Trigger signals (offer the question when ANY are present):

• Verbs: refactor, reuse, extract, consolidate, migrate, rename, split, merge, deduplicate, unify, abstract, move, replace
• Implementation nouns paired with an action: component, hook, utility, service, module, layout, route, page, store, context, provider
• Explicit file paths, component names (e.g., Sidebar), or function/hook names

If triggered, use AskUserQuestion:

"Your request touches code (e.g., refactoring/reuse). Do you want me to dive into the codebase and include relevant details in the ticket?"

Options:

• Yes, research — Explore the codebase and add findings to the description
• No, skip — Proceed without code research

If Yes, spawn an Explore agent (via the Agent tool with subagent_type: Explore) to find:

• Current location(s) of the subject code (file paths with line references)
• Similar or duplicated patterns elsewhere in the codebase
• Files/components/consumers that would be affected by the change
• Existing conventions or abstractions to align with

Keep the summary concise — bullet points with path:line references, not prose walls. These findings feed into 2b (title sharpness, description clarity, and a new Details section in the body).

2b. Draft the Proposal

Based on the user's input, answers so far, and any code research findings:

1. Propose a title — concise, under 80 characters. If research was done, make it specific (e.g., "Refactor Sidebar for reuse in ModelList, AdapterList, UserList").

2. Propose a summary — structured as:

   • A short description paragraph (1-3 sentences explaining what and why)
   • Key points as a bullet list extracting the important details
   • For bugs: also include drafted "Actual result" and "Expected result" if enough context exists
   • If code research was done: a Details section listing current location, affected files, duplicated patterns, and recommended scope (with path:line references)

3. List all labels that will be auto-assigned based on the issue type and any labels already determined from context (e.g., bug, Design Required if obvious from description). Mark labels that will be asked about later as "TBD".

Present the proposal to the user and ask for confirmation:

"Here's what I've drafted from your input:

Title: <proposed title>

Summary:

Key points:

• <point 1>
• <point 2>
• ...

Labels (planned):

• label1 (auto)
• label2 (auto)
• Priority — TBD
• Severity — TBD
• ...

Want to use this, or change anything?"

Options:

• Use as-is — proceed with this draft
• Edit — I want to change something

If Edit: ask what they want to change, revise, and re-confirm.

The proposed summary becomes the main content of the description/body field.

IMPORTANT: If the user approves the draft ("Use as-is"), do NOT re-ask fields that were already covered in the draft. In Step 3, only ask for fields that are missing or were not part of the proposal. For example, if the draft already includes "Actual result" and "Expected result" for a bug, skip those questions and only ask for remaining fields (version confirmation, steps to reproduce, severity, etc.).

---

Step 3 — Type-Specific Fields

Before asking individual fields, if the draft from Step 2 already covers some fields, ask the user:

"The draft already covers some details. How do you want to proceed?"

Options:

• Use draft, fill remaining — Accept drafted fields, only ask for missing ones (version, severity, etc.)
• Provide details now — Go through each field interactively
• Use draft as-is — Skip all type-specific questions, use drafted content for all fields and fill missing fields with reasonable defaults or "No response"

Gather information interactively based on the issue type. Ask each field as a separate question using AskUserQuestion (open-ended, no preset options) unless a dropdown/choice is specified.

Bug

1. Version: Auto-detect by reading the version field from the root package.json. Confirm with user:

   "The current version is <version>. Is this the version where you see the bug?"

2. Steps to reproduce: Ask:

   "How do you reproduce this bug? Provide step-by-step instructions."

3. Actual result: Ask:

   "What happens currently? (the broken behavior)"

4. Expected result: Ask:

   "What should happen instead?"

5. Additional information (optional): Ask:

   "Any additional context? (screenshots, logs, browser info — or skip)"

6. Severity: Use AskUserQuestion with options:
   • Critical — Severe issue impacting core functionality, requires urgent resolution
   • Major — Affects many users or key components
   • Minor — Minor issue with limited impact
   • Low — Low issue with little to no critical impact

Feature

1. Description: Ask:

   "Describe the feature. What should it do and why is it needed?"

2. Related issues (optional): Ask:

   "Are there any related issues? (paste issue numbers/URLs or skip)"

Task

Covers engineering work (refactor, reuse, tech debt, cleanup, test/build improvements) AND infrastructure changes (env var, secret, config, deployment setting).

Step 3.0 — Infra sub-classification

Before gathering fields, detect whether this Task is an infrastructure change.

Infra signals (auto-apply ops-request label when ANY are present):

• Keywords: environment variable, env var, secret, credential, config change, deployment, LOG_LEVEL, NODE_ENV, prod / uat / development, Kubernetes, Helm, CI/CD, pipeline
• The user invoked /create-ticket infra: … (args prefix forces ops-request label)
• The description only describes a runtime/config change with no code change

If infra signals are clearly present → auto-apply ops-request label (tell the user), go to the infra field set below. If genuinely ambiguous → use AskUserQuestion:

"This looks like a Task. Is it an infrastructure change (env var, secret, config, deployment)?" Options: Yes, infra (adds ops-request label) / No, general task

General Task fields (no ops-request label):

1. Description: Ask:

   "Describe the task. What needs to change and why? (problem/motivation)"

2. Acceptance criteria (optional): Ask:

   "List acceptance criteria or sub-tasks as checklist items (one per line), or skip." Format each as - [ ] <item> in the body.

3. Related issues (optional): Ask:

   "Are there any related issues or PRs? (paste numbers/URLs or skip)"

Infra Task fields (ops-request label auto-applied):

1. Change type: Use AskUserQuestion with options:
   • environment variable
   • secret
   • config change
   • extra configuration
   • other
2. Target environment: Use AskUserQuestion with options:
   • development
   • uat
   • production
   • all
3. Task list: Ask:

   "List the changes needed as checklist items (one per line)." Format each as - [ ] <item> in the body.

4. Context / reason (optional): Ask:

   "Why is this change needed? Link to a ticket, incident, or PR if applicable. (or skip)"

---

Step 4 — Priority

Use AskUserQuestion:

"What priority level?"

Options (map to the repo's P# labels — note the exact label text includes the en dash –):

• Critical — Feature or system completely broken for all users, no workaround → P1 – Critical
• High — Broken for a specific segment; workaround exists but inconvenient → P2 – High
• Medium — Secondary flow/UX degraded, reasonable workaround → P3 – Medium
• Low — Cosmetic or edge-case, trivial/no workaround needed → P4 – Low

---

Step 5 — Conditional Labels

Evaluate the description and context gathered so far. For each label below:

• If it is 100% clear from context that the label applies → auto-assign it and inform the user
• If it is uncertain → ask the user directly

Design Required

Auto-assign yes if the user explicitly mentions: new page, new UI component, layout change, redesign, UX change, new visual element.

Auto-assign no (skip the question entirely) when the context clearly has no design implications — e.g., crashes, errors, backend issues, config changes, refactoring, performance bugs, infra tasks.

Only ask the user when it's genuinely ambiguous (e.g., "improve the user list" — could be UX or just data):

"Does this issue require design work?" Options: Yes / No

Security Impact Analysis

This repo has no SIA-* labels, so security impact is not captured via a label. If the user mentions a security-relevant surface (authentication, authorization, tokens, secrets, passwords, permissions, session, credentials, encryption, PII), note "Security impact: needs review" in the issue body instead of applying a label. Do not attempt to add an SIA-required / SIA-not required label — creation will fail (label not found).

Area label (optional)

The repo has area labels — analytics-2.0, eval, cli, ui-kit. When the issue clearly belongs to one area, offer to add the matching label (e.g. an Analytics feature → analytics-2.0). Only apply when confident; otherwise skip.

---

Step 6 — Assignee

A ticket MUST have an assignee — never leave one unassigned (tickets without owners get lost). Default to the creator (@me).

Use AskUserQuestion:

"Assign to you (the creator) by default, or someone else?"

Options:

• Me (default) — Assign to the creator (@me)
• Someone else — Ask for a GitHub username

If Someone else, ask:

"GitHub username to assign?"

---

Step 7 — Preview

Show the user a formatted preview of the entire issue:

══════════════════════════════════════════
ISSUE PREVIEW
══════════════════════════════════════════

Title:    <title>
Type:     <Bug/Feature/Task> (reflected via labels)
Labels:   <comma-separated list — includes `ops-request` if this is an infra change>
Assignee: <@me (creator) / username>

──────────────────────────────────────────
BODY:
──────────────────────────────────────────

<formatted body matching template structure>

══════════════════════════════════════════

Use AskUserQuestion:

"Create this issue?"

Options:

• Create — Create the issue now
• Edit — I want to change something
• Cancel — Don't create the issue

If Edit: Ask what they want to change, update it, and show the preview again. If Cancel: Stop and confirm cancellation.

---

Step 8 — Create

Build and execute the gh command:

gh issue create \
  --title "<title>" \
  --body "<body>" \
  --label "<label1>,<label2>,..." \
  --assignee "<@me|username>"

Notes:

• --type is not supported by gh issue create. Issue type (Bug/Feature/Task) is conveyed via labels (bug, enhancement, or task-specific labels) — there is no separate type flag.
• There is no separate "Infra Task" type — infrastructure work is a Task with the ops-request label auto-applied and the infra-specific body structure (change type, target environment, task list).
• --assignee is always set (default @me), so every ticket lands with an owner rather than in an unassigned backlog.
• The <details-section> placeholder in the templates below is the code-research findings from Step 2a. If research was NOT performed, omit the entire ### Details heading and its content — do not leave an empty section.
• The body must be formatted to match the GitHub issue template output format:

Body formats: references/issue-bodies.md holds the exact per-type body structure (Bug, Feature, Task, infra Task). Read it before composing the body — the headings must match the repository's issue templates verbatim.

After successful creation, display the issue URL returned by gh.

---

Label Reference

references/labels.md holds the verified label table for this repo. gh issue create fails hard on an unknown label, so confirm against it — or gh label list --limit 300 — before applying.

Guardrails

• Creating an issue is outward-facing and awkward to undo: show the preview and get explicit confirmation first, every time.
• Only apply a label you are certain of. gh issue create fails outright on a label this repo doesn't have, and a wrong-but-valid label quietly misroutes the ticket.
• When uncertain about ANY label, ask directly — don't guess
• Always use the exact label names as listed in the Label Reference table
• The confidential information checkbox is always pre-checked in the body
• If gh CLI fails, show the error and suggest the user check their auth (gh auth status)