The Windows process launcher no longer needs Bash/PowerShell mediation, and the event-source tests now exercise CLI boundaries through an injectable executor instead of generated scripts. This keeps production behavior on real external commands while letting tests model long-running process streams and kill semantics in Dart.
Constraint: Tests must not mock CLIs by writing generated Dart or shell scripts and executing them.
Rejected: Use generated Dart script fixtures | still depends on subprocess execution for mocks.
Rejected: Use only mocktail mocks for process handles | stream, kill, and exit behavior are clearer as a small stateful fake.
Confidence: high
Scope-risk: moderate
Directive: Keep long-running CLI test behavior behind ExternalCommandExecutor rather than reintroducing script fixtures.
Tested: dart analyze; dart run tool\\validate_gherkin_test_format.dart; dart test
Not-tested: Real gh/gosmee webhook forwarding against live services
Stop overriding responder no_reply decisions for mention-triggered comments so automation mentions do not loop indefinitely. Update the default prompt to encourage mention handling without forcing replies and add regression coverage for no_reply mention flows.
Add glab-backed GitLab issue polling, login lookup, and comment posting so GitLab projects can be handled alongside GitHub, Gitea, and OpenProject. Also wire GitLab through SCM selection, config defaults, docs, and examples so issue #35 can be configured end to end.
GitHub rejects webhook creation when gh api serializes active as a string. Use a typed field so gh sends a real boolean, and lock the gh/gosmee integration test harness to catch regressions.
Replace the plain logging backend with a color-capable logger adapter so CLI output is easier to scan, and add richer gosmee/webhook-forward diagnostics so tracker and forwarding failures can be debugged from logs.
* feat: add github webhook-forward event source for #12
Add a channel-based GitHub issue source so long-running sessions can react to issue and issue_comment webhooks without polling. Keep --once on polling so backlog reconciliation still works and cover the new config/runtime path with tests.
* refactor: extract polling issue event source (#10)
Decouple tracker event discovery from issue processing so channel-based sources can plug into the runtime later without rewriting reply handling.
Rename the provider-aware tracker client and scope since cutoffs per provider so review-driven cleanup improves readability and avoids extra tracker fetches.
fix: Export workspace manager in code work spawner
feat: Enhance CLI responder to handle JSONL events
feat: Implement job status enumeration in database
feat: Add logging capabilities to IssueAssistantApp
insleker