github-issues

name	github-issues
description	Create GitHub issues from user requirements, a PRD document, or conversation context, and add them to a GitHub Project with priority, size, and type fields.
argument-hint	[source-file-path] [--project <number>]

GitHub Issues

Create well-structured GitHub issues from any input — a PRD document, a requirements description in chat, feature ideas, or a task breakdown — then add them to a GitHub Project board with Priority, Size, and Type metadata.

Input

The user may provide:

• A path to a document (PDF, Markdown, text) via $ARGUMENTS
• Requirements described directly in the conversation
• A mix of both

If $ARGUMENTS contains a file path, read that file. Otherwise, work from what the user described in chat. If no project number is given, detect it or ask.

Step-by-step process

1. Understand the requirements

• If a file path is provided in $ARGUMENTS, read and analyze it
• If requirements are described in conversation, use that context
• Extract: features, user flows, screens, technical requirements, integrations
• Identify distinct work items that can each be assigned to one developer
• Ask clarifying questions if the scope is ambiguous

2. Load or create project configuration

Look for a cached config file at this path (colocated with this skill):

.claude/skills/github-issues/.github-project-config.json

If the file exists, read it and use all IDs directly — skip all GitHub API discovery calls.

If the file does NOT exist, run the full discovery flow below, then save the results to that path so future runs are instant. Also ensure .github-project-config.json is in the repo's .gitignore (the pattern **.github-project-config.json or the exact path).

Discovery flow (first run only)

a) Detect repo info:

gh repo view --json name,url,owner

Determine if the owner is a user or organization (check the owner.id prefix — O_ = org, U_ = user, or use gh api users/<login> --jq .type).

b) List labels:

gh label list

c) Find the project:

gh project list --owner <owner>

If a project number was provided in args, use it. If multiple projects exist, ask the user. If only one, use it automatically.

d) Query project fields (Priority, Size, Status) and their option IDs:

gh api graphql -f query='{
  <ownerType>(login: "<owner>") {
    projectV2(number: <N>) {
      id
      fields(first: 30) {
        nodes {
          ... on ProjectV2SingleSelectField {
            name
            id
            options { id name }
          }
        }
      }
    }
  }
}'

e) Query repo-level issue types:

gh api graphql -f query='{
  <ownerType>(login: "<owner>") {
    repository(name: "<repo>") {
      issueTypes(first: 20) {
        nodes { id name description isEnabled }
      }
    }
  }
}'

f) Save config: Write all collected data to .github-project-config.json using the format in the "Config file format" section. Only include fields that actually exist on the project — omit priority, size, or issueTypes if the project doesn't have them.

3. Plan the issues

Before creating anything, present the user with a summary table:

#	Title	Labels	Priority	Size	Type

Group issues by area (Backend, Web, Mobile, Database, Infra, Docs, etc.).

Ask the user to confirm before proceeding with creation. Let them adjust titles, priorities, sizes, or remove issues.

4. Create labels (if needed)

Check the labels array in the config. Only create labels that don't already exist:

gh label create "<name>" -c "<color>" -d "<description>"

After creating new labels, update the config file's labels array.

Do NOT create priority labels (P0/P1/P2) — use project fields instead.

5. Create issues

For each issue, use gh issue create with a heredoc body:

gh issue create --title "<title>" --label "<label1>,<label2>" --body "$(cat <<'EOF'
## Description
<clear description of what needs to be built>

## Requirements
<bullet list of specific requirements>

## Implementation notes
<relevant technical details, existing code to reuse, file paths>

## Files
<list of files to create or modify>
EOF
)"

Guidelines for issue content:

• Title: concise, action-oriented (e.g., "Student auth: OTP send & verify flow")
• Description: what to build, not how (leave implementation to the developer)
• Include relevant API endpoints, screen descriptions, or data models
• Reference existing code/files when applicable
• Keep each issue self-contained and assignable to one developer

6. Add issues to the project

Use project.number and repo.owner from the config:

gh project item-add <project.number> --owner <repo.owner> --url <issue-url>

Collect the returned item IDs for field assignment in the next step.

7. Set project fields (Priority, Size, Type)

Use IDs directly from the config — no GraphQL lookups needed. Skip any field that doesn't exist in the config.

For Priority and Size, use gh project item-edit:

gh project item-edit --project-id <project.id> --id <item-id> \
  --field-id <fields.priority.id> --single-select-option-id <fields.priority.options[value]>

For issue types (Feature, Task, Bug), write the GraphQL mutation to a temp file to avoid shell escaping issues:

cat > /tmp/gql_query.txt << 'GQL'
mutation($issueId: ID!, $typeId: ID!) {
  updateIssueIssueType(input: {issueId: $issueId, issueTypeId: $typeId}) {
    issue { number }
  }
}
GQL
gh api graphql -F query=@/tmp/gql_query.txt -F issueId=<issue-node-id> -F typeId=<issueTypes[value]>

To get the issue node ID for type assignment:

gh api graphql -f query='query { repository(owner: "<repo.owner>", name: "<repo.name>") { issue(number: <N>) { id } } }'

If the project scope is missing for item-edit, inform the user to run gh auth refresh -s project and retry.

8. Verify and report

After all issues are created:

gh issue list --limit 100 --state open

Present a final summary to the user with:

• Total issues created, grouped by area
• Priority breakdown (P0/P1/P2 counts)
• Size breakdown (XS/S/M/L/XL counts)
• Link to the project board

Config file format

The .github-project-config.json file should have this structure. Only include fields that exist on the project — all sections except repo are optional:

{
  "repo": {
    "name": "repo-name",
    "owner": "org-or-user",
    "ownerType": "organization | user"
  },
  "project": {
    "number": 1,
    "id": "PVT_..."
  },
  "fields": {
    "status": {
      "id": "PVTSSF_...",
      "options": { "Backlog": "id", "Todo": "id", "In Progress": "id", "Done": "id" }
    }
     
       
              
    
     
       
                  
    
  
   
     
     
     
  
      

Issue sizing guide

Size	Scope
XS	Simple CRUD, single config, small copy change
S	One endpoint or one screen, straightforward logic
M	Multiple endpoints or screens, moderate complexity
L	Complex feature spanning multiple files/systems
XL	Large cross-cutting feature, significant architecture

Priority guide

Priority	Meaning
P0	Must-have for launch, blocks other work
P1	Important, build after P0 items are done
P2	Nice to have, can ship without it

Type assignment guide

Type	When to use
Feature	New user-facing functionality, API endpoints, UI screens
Task	Setup, configuration, CI/CD, documentation, testing infrastructure
Bug	Only if fixing existing broken behavior

Important notes

• Always check for .github-project-config.json first — skip all discovery if it exists
• On first run, create the config file automatically after discovery so future runs are instant
• Ensure .github-project-config.json is gitignored (it contains project-specific IDs, not secrets, but is environment-specific)
• Always ask the user to confirm the issue plan before creating
• Never create priority labels — use project fields instead
• Check for existing labels before creating new ones
• Use heredocs for issue bodies to preserve formatting
• For GraphQL mutations, write the query to a temp file and use -F query=@file to avoid shell escaping issues
• Batch operations where possible for efficiency
• If the repo has no project, ask if the user wants issues only (skip project-related steps)
• If gh auth scopes are insufficient, tell the user which scope to add
• If the config seems stale (e.g., field IDs fail), delete it and re-run discovery