Orchestrate parallel AI coding agents across a polyrepo codebase from a single devcontainer or vscode workspace.
- Run one Claude Code prompt across every repo in the workspace, in parallel
- Presets for repo-wide validation, a repo-wide security audit, and a diff-scoped review of untrusted PRs
- One multi-root VS Code workspace with a terminal task per repo, opened automatically
- Each sibling repo's own devcontainer setup replayed inside the host container
- One repo list (
config/repos.conf) drives cloning, workspace generation and runs - A read-only GitHub-side estate sweep (
scripts/estate-status.sh): open PRs/issues, Dependabot/CodeQL alerts, CodeFactor and CI state, across every repo an owner has, scoped to--scope public(default),private, orall— private repos are only ever shown as maskedprivate-NNlabels, and mentions of their names in other repos' titles are scrubbed too (add--redact-from FILEfor repos a run can't discover on its own, e.g. a renamed or deleted private repo)
./scripts/cc-parallel.sh --preset validate
./scripts/cc-parallel.sh --preset security-all # repo-wide audit (all repos)
./scripts/cc-parallel.sh --preset security-pr # diff-scoped, untrusted PRs
./scripts/cc-status.sh
./scripts/estate-status.sh --scope public # GitHub-side estate sweep; private|all also availableRepos: edit config/repos.conf. Credentials: set GH_PAT as Codespace secret.
Setup lifecycle (make setup_all on create, make setup_repos on attach): see
Codespaces.
A VS Code multi-root workspace only runs the host container's devcontainer lifecycle, and an agent session works one repo at a time. For teams running Claude Code (or other AI agents) across many repos at once, polyforge replays each repo's setup in one container and fans a prompt out to all of them.
- Codespaces — rebuild, secrets, management
- Cross-repo setup — auth, sandbox, settings
- Cloud workflows — remote execution
- Sandbox friction — known issues, mitigations
- Engineering-practice baseline — dev-loop presets mapped to the Startup CTO Handbook
