Multi-Repo Sessions

Multi-repo sessions allow Bucky to work across multiple repositories simultaneously. This is useful for tasks that span services — like updating a shared API contract, coordinating a migration, or investigating a cross-service issue.

How it works

1. The agent’s run_multi_repo_session tool dispatches a repository_dispatch event with type bucky-multi-repo-session to the buck-bronson repo itself
2. The bucky-multi-repo.yml workflow clones all specified repos into repos/<name>/ subdirectories
3. Claude runs with multi-repo-specific prompts that are aware of the multi-repo layout
4. Results are reported back via the same callback pattern as single-repo sessions

Workflow details

The bucky-multi-repo.yml workflow:

• Clones repos from client_payload.repos (an array of owner/repo strings)
• Sets up Node.js and Go toolchains without running package installation
• Registers all plugins (common, web, golang, LSPs) but only injects SOURCEGRAPH_TOKEN
• Uses prompts from .github/actions/bucky-execute/prompts/ with additional placeholders:

Placeholder	Value
{{ISSUE_REPO}}	The repo where the GitHub issue lives (always buck-bronson)
{{REPO_LIST}}	Comma-separated list of repos
{{REPO_PATHS}}	Paths where repos are cloned (repos/<name>/)

Key differences from single-repo sessions

Aspect	Single-repo	Multi-repo
Runs in	Target repo	buck-bronson
GitHub issues	In target repo	In buck-bronson
Package install	Yes (per preset)	No
Secrets	Per-preset tokens	Only SOURCEGRAPH_TOKEN
Dispatch type	bucky-session	bucky-multi-repo-session

Caution

Multi-repo sessions do not install dependencies for individual repos. Claude has access to the source code but cannot run project-specific build/test commands unless the tools are already globally available in the runner image.