| name | do |
| description | Step-based workflow for implementing an issue in a multi-repo workspace. Handles research, planning, shallow cloning, agent teams, implementation, QA, and pull requests. |
| disable-model-invocation | true |
/do — Implement an Issue Across Multiple Repos
Step-based workflow for implementing an issue end-to-end in a parent repo that orchestrates work across multiple child repos via on-demand shallow cloning.
Input
Accepts:
- Issue URL:
https://your-tracker.example.com/browse/PROJ-123
- Issue ID:
PROJ-123
- Just the number:
123 (assumes the configured project prefix)
Steps
1. Fetch & Understand the Issue
- Use your issue tracker tool to view the issue
- Read title, description, acceptance criteria, comments
- Transition to "In Progress"
2. Research & Planning
- Use a code search tool (e.g., SourceBot's
codesearch MCP) to find what code is involved and where it lives
- IMPORTANT: Use
AskUserQuestion to clarify ambiguities before proceeding. It is ok to over-ask to clarify details for the spec!
- Read relevant code to understand existing patterns before making changes
3. Update the Issue
- Update the issue title and description with refined requirements, acceptance criteria, and technical plan before implementing
- This is critical — the issue must accurately reflect what will be built
4. Shallow-Clone Required Repos
Only clone the repos you need. Use --depth 1 for fast, minimal clones:
git clone --depth 1 https://github.com/org/repo-name.git
Do NOT clone all repos upfront — only the ones identified as relevant during planning.
This keeps initialization fast and reduces noise while grepping and globbing. Because we use --depth 1, only the latest commit is fetched — no full history.
5. Spin Up Agent Team
Spin up an Agent Team with at least one agent per repo being modified:
- Each agent works in its own cloned repo directory! YOU MUST INITIALIZE THE AGENT IN THE CORRECT DIRECTORY!
- Each agent creates a feature branch, implements changes, runs tests, and commits
- The lead agent coordinates, reviews progress, and handles cross-repo concerns
6. Implement
- Prefer red-green TDD — write tests first when possible
- Follow existing architectural patterns in each repo
- Run pre-commit hooks and tests in each repo before committing
7. QA
There should be another QA agent per repo being worked on that should run a code review against the PR/MR as recommended by the repo's agentic documentation, run tests, and run the services to make sure they're working as expected.
And there should be an over-arching integration test agent that runs integration tests across all repos being worked on with help via communicating via each of the per-repo subagents.
Do NOT skip this step — it is critical to the success of the project.
8. Create Pull Requests
- Create PRs/MRs via your VCS tool
- Title: always prefix with the issue key —
[PROJ-123] Descriptive title
- Description: comprehensive — include summary, what each agent did, QA steps, and overview of all changes
- NEVER mark the issue as "Done" — that's for human review after merge
There should be one PR/MR per child repo worked on!
There should always be a PR/MR in the parent repo itself that serves as a reference for the others: the description should link to the other PRs/MRs and give a high-level overview of the changes.