ai-worker-testing

Layout and running

• Specs live in spec/{models,services,jobs,requests}, mirroring app/. Helpers: spec/rails_helper.rb, spec/spec_helper.rb.
• Always run with the environment set explicitly: RAILS_ENV=test bundle exec rspec (or a single file/line). Never rely on the default environment — it is development, which is real data. Test DB is storage/test.sqlite3; prepare with RAILS_ENV=test bin/rails db:test:prepare.
• Transactional fixtures are on (use_transactional_fixtures = true); records created in an example roll back automatically.
• CI runs the same commands in the test job (.github/workflows/ci.yml) and in config/ci.rb (bin/ci).

Data safety

• Never touch development data. No destroy_all / delete_all / truncate "clean slate" cleanup — rely on transactions, or delete only the exact records you created.
• For Project#local_directory use Dir.mktmpdir and remove only that dir in an after hook, guarded by start_with?(Dir.tmpdir) (see spec/models/project_spec.rb).

No real external calls

• Never invoke real claude, gh, or git against the network. Stub Open3.capture3, Process.spawn, AiCli::Runner.start / .poll.
• For Git::BranchService, use a scratch local git repo created in a tmpdir (no remote), and remove only that tmpdir afterwards.

Design

• Prefer extracting pure logic into small POROs (e.g. AiCli::Output, Prompts::StagePrompt) and unit-testing them directly, over testing through jobs or subprocesses.