Instant Software Factories

This project provides a ready-made bootstrap that turns a GitHub repository into an automated "software factory" powered by scheduled Claude Code routines. It requires a repo with Actions and enforceable branch protection and a Claude plan that supports cloud routines. The setup prompt and BOOTSTRAP.md inspect the codebase, CI, branch protection, and credential history; install gh if needed; create a minimal CI workflow when absent; and produce a tested setup pull request that preserves existing safeguards. Routines run under the connected GitHub identity, so their PRs require a separate human approver; production schedules remain disabled until the setup PR is merged. Budgeting for both agent calls and GitHub Actions minutes is emphasized.

Operation is organized into standing-agent roles (QA, Performance, CI, Architecture, Implementation, PR Review, Merge, and Manager) that run on schedules, read repository state, choose work, create issues and PRs, and record evidence. The manager creates and configures routines (create_trigger plus local RemoteTrigger updates), prioritizes backlog, tracks cost per merged PR, and adjusts schedules to match CI and reviewer capacity. The system keeps recurring attention on tests, performance, architecture, and integration while human approval gates changes. Main limitations are CI capacity and platform reliability: agents can produce queues, conflicting fixes, or reviewer bottlenecks, so throughput must be tuned to available validation and human review.