create

name	create
description	Create well-researched GitHub issues with immediate refinement. Use when user says "create issue", "new issue", "file a bug", "open an issue", "report a problem", "request a feature", or wants to create a GitHub issue from a rough idea.

Create

Creates a new GitHub issue from a rough idea, immediately refining it into a well-structured issue with deep research.

Critical Rules

• Read cross-cutting behaviors first: skills/shared/references/cross-cutting.md
• Research before writing — use every available tool to gather context
• Always search for duplicates before creating
• Always preview the issue with the user before submitting
• NEVER create priority labels — see skills/shared/references/label-taxonomy.md
• Add a creation comment summarizing research sources and decisions

Prerequisites Check

gh auth status

If this fails, tell the user to run gh auth login first.

Workflow

Phase 1: Understand the Intent

Parse the user's request. Identify:

• Is this a bug report, feature request, task, or something else?
• What problem does it solve or what value does it add?
• Are there specific details (error messages, file paths, steps to reproduce)?

Phase 2: Deep Research

Use every available tool to gather as much context as possible. The goal is to create an issue that is immediately actionable — not a placeholder.

2a. Codebase Research

Scan the local codebase for relevant context:

• Search for related code — find files, functions, classes related to the issue topic
• Read relevant files — understand the current implementation
• Check for TODOs/FIXMEs — existing markers for known issues
• Review recent commits — has this area been recently changed?

# Recent changes in relevant area
git log --oneline -20 -- "path/to/area"

2b. Web Research

Search the web for relevant context — this is critical for well-informed issues:

• Error messages — search for exact error text to find known issues, solutions
• Library/framework issues — check upstream bug trackers for related problems
• Best practices — search for recommended approaches to the problem
• Documentation — find official docs for relevant APIs, libraries, tools

Use WebSearch and WebFetch to gather information. Always verify information against the actual codebase.

2c. Available Tools and Servers

Use any MCP servers and tools available in the current session:

• Asana/project management — check for related tasks or initiatives
• Documentation tools — search internal docs for context
• API tools — verify endpoint behavior, check schemas
• Any other connected service that might have relevant information

2d. GitHub Research

# Search for duplicate or related issues
gh issue list --search "keyword1 keyword2" --state all --json number,title,state,labels --limit 20

# Check closed issues — has this been reported and closed before?
gh issue list --search "keyword" --state closed --json number,title,closedAt --limit 10

# Check PRs that might be related
gh pr list --search "keyword" --state all --json number,title,state --limit 10

# Check existing labels
gh label list --json name,color,description --limit 100

If duplicates are found, tell the user and ask whether to:

• Add a comment to the existing issue instead
• Create a new issue with a cross-reference
• Abort

Phase 3: Draft the Issue

Based on all research, compose a well-structured issue.

For bug reports:

## Problem

[Clear description of the defect]

## Steps to Reproduce

1. [Step 1]
2. [Step 2]
3. [Step 3]

## Expected Behavior

[What should happen]

## Actual Behavior

[What actually happens, include error messages]

## Context

- [Relevant code: `path/to/file.py:42`]
- [Environment details if relevant]
- [Related: #N, upstream-repo#M]

## Acceptance Criteria

- [ ] [Specific testable criterion]
- [ ] [Edge case to handle]

For feature requests / user stories: Read skills/refine/references/user-story-guide.md for the full template, then apply:

## User Story

As a [user type],
I want [goal],
So that [benefit].

## Context

[Background, motivation, research findings, links to docs or discussions]

## Acceptance Criteria

### Scenario: [happy path]
- Given [precondition]
- When [action]
- Then [expected result]

### Additional criteria
- [ ] [Criterion]
- [ ] [Non-functional requirement]

## Out of Scope

- [What this does NOT cover]

## Technical Notes

- [Relevant code areas: `path/to/module/`]
- [Dependencies or constraints discovered during research]
- [Links to relevant documentation]

## Related Issues

- Related to #N
- See also #M

Phase 4: Validate

Before presenting to the user, check the draft against:

1. INVEST criteria (read skills/refine/references/user-story-guide.md):

   • Independent, Negotiable, Valuable, Estimable, Small, Testable
   • If the issue is too large, suggest splitting into an epic with sub-issues

2. Completeness — does the issue have enough context to act on?

   • For bugs: can someone reproduce it from the description alone?
   • For features: are the acceptance criteria specific and testable?

3. Labels — suggest appropriate labels from the existing set

Phase 4b: Review

Before finalizing any issue creation, delegate to the issue-reviewer agent to validate the proposed changes. Present the review results to the user. If the review fails, fix the issues before proceeding.

Phase 5: Preview and Create

Present the complete draft to the user:

• Title
• Body
• Suggested labels
• Suggested assignee (if obvious)
• Any related issues found

Wait for user approval or feedback. Iterate if needed.

gh issue create \
  --title "Title" \
  --body "Body with markdown" \
  --label "feature" \
  --assignee @me

After creation:

• Report the issue number and URL
• If the issue is large enough to be an epic, offer to decompose it using the refine skill

Troubleshooting

gh not found:

• Tell the user to install it: brew install gh (macOS) or see https://cli.github.com

Not in a git repository:

• Ask the user to navigate to a repository or specify --repo owner/name

Issue creation fails:

• Verify the repo has issues enabled: gh repo view --json hasIssuesEnabled
• Check that labels exist before referencing them in create

No web search results:

• Fall back to codebase-only research. Note in the issue that external research was limited.