Skip to content
Agent Orchestration
Command

/create-issue

Create a comprehensive, well-structured GitHub issue with embedded agent instructions for autonomous PR lifecycle management.

BOOST
From plugin
metaswarm
42116 skills19 agents16 commands2 hooks
Install
$ npx -y skills add dsifry/metaswarm --agent claude-code

How it fires

How this command gets triggered: by you, by Claude, or both.

  • Fires itselfClaude auto-loads it when your prompt matches the work.
  • You can call itInvoke it directly when you want it.
  • Slash command/create-issue

Context preview

What this command does when you run it.

Create a comprehensive, well-structured GitHub issue with embedded agent instructions for autonomous PR lifecycle management.

Command definition

create-issue.md

Create GitHub Issue

Create a comprehensive, well-structured GitHub issue with embedded agent instructions for autonomous PR lifecycle management.

Usage

/create-issue [type] [description]

**Arguments** (optional - will prompt if not provided):

  • `type`: `feature`, `bug`, or `refactor`
  • `description`: Brief description of the issue

Examples

# Interactive mode (prompts for all details)
/create-issue

# With type specified
/create-issue feature

# With type and description
/create-issue bug "Login fails when email contains plus sign"

What This Does

1. **Gathers context** through interactive questions 2. **Generates comprehensive issue** with all required sections:

  • Problem/Feature specification with clear scope
  • Technical specification (types, schemas, APIs)
  • TDD Implementation Plan with test-first cycles
  • Error handling matrix
  • Files to create/modify
  • Acceptance criteria checklist
  • **Agent Instructions** for complete PR lifecycle

3. **Creates GitHub issue** with appropriate labels 4. **Returns issue number** for tracking

Steps

1. **Activate the create-issue skill**: Load and follow the create-issue skill definition.

2. **Gather information** through the interactive flow:

  • Issue type (Feature/Bug/Refactor)
  • Brief description
  • Complexity estimate
  • Related files/services

3. **Generate issue content** using appropriate template

4. **Create issue** via `gh issue create`

Why Agent Instructions Matter

Every issue includes an "Agent Instructions" section that ensures:

  • **Complete lifecycle**: From implementation through merge
  • **All comments addressed**: Including trivial/nitpicks and out-of-scope
  • **Individual responses**: Every thread gets a reply
  • **Iteration until done**: Wait for reviews, iterate, don't auto-resolve
  • **Issue creation for deferred work**: > 1 day = new issue

This makes issues self-contained work packets that agents can execute autonomously while maintaining quality standards.

Issue Structure

All generated issues follow this structure:

## Summary
## Problem / Bug Report
## Architecture Decision (features/refactors)
## Technical Specification
## TDD Implementation Plan / Test Cases
## Error Handling
## Implementation Phases
## Files to Create/Modify
## Acceptance Criteria
## Agent Instructions  <-- CRITICAL for autonomous work
## Related Issues
## References

Labels

Issues are automatically labeled based on type and complexity:

| Type | Labels | | -------- | ----------------------------------- | | Feature | `enhancement`, `complexity:[level]` | | Bug | `bug`, `priority:[level]` | | Refactor | `refactor`, `complexity:[level]` |

Related Commands

  • `/handle-pr-comments` - Handle PR review feedback
  • `/pr-shepherd` - Monitor PR through to merge
  • `/create-pr` - Create comprehensive PR
  • `/start-task` - Start working on an existing issue
Read more
Ships withmetaswarm

A self-improving multi-agent orchestration framework for Claude Code, Gemini CLI, and Codex CLI.

Get the whole plugin

Other commands on metaswarm.