These issues are ready to publish when the GitHub integration has issue-write permission. The titles, labels, scope, and acceptance criteria are intentionally concrete so each issue is a real entry point rather than a placeholder.
Labels: good-first-issue, area: cli, enhancement
Audit settings and onboarding validation. Replace generic failures with actionable, secret-safe guidance for a missing key, unknown provider, unknown model, and unreachable local endpoint. Valid configurations must remain unchanged and focused unit tests must cover the new messages.
Labels: good-first-issue, area: packaging, os: linux
Run the documented installation, --help, --version, first-run setup, and one stub-backed task on a clean supported Ubuntu environment. Add an onboarding report under docs/onboarding_reports/, include reproduction steps for every failure, and update confirmed documentation or script gaps.
Labels: help-wanted, area: engine, provider-bug
Test OpenAI, Anthropic, Google, OpenRouter, Ollama, and LM Studio where access is available. Report provider, model, date, OS, model listing, streaming, structured tool calls, cancellation, and a minimal agent task. Label untested cells honestly and split confirmed bugs into focused issues.
Labels: help-wanted, provider: lmstudio, area: engine
Validate model discovery, streaming, tool calling, cancellation, a read-only task, a small edit/test task, and offline-endpoint guidance against a current LM Studio server. Record exact versions and add a sanitized known-good settings example.
Labels: documentation, os: macos, area: packaging
Test a clean supported Python environment, pipx, PATH troubleshooting, optional local-model setup, and first launch. Publish copy-pasteable commands, a sanitized onboarding report, and only verified Apple Silicon or Intel differences.
Labels: help-wanted, evaluation, area: engine
Add a self-contained task with independently defined tests, a pinned prompt, explicit budgets, and reporting for runtime, model, tokens/cost, tool outcomes, and human intervention. Keep agent FINISHED state distinct from external acceptance-test compliance.
Labels: help-wanted, ux, area: cli, enhancement
Make hosted-versus-local choices, credentials, endpoint checks, and a safe test action understandable from a clean state. Preserve keyboard-only navigation and narrow-terminal usability; back and cancel paths must not leave partial secrets or invalid settings.
Labels: help-wanted, performance, area: engine
Create a repeatable long-session benchmark for compaction duration, prompt size before and after, peak RSS, cache behavior, and retrieval overhead at multiple context-pressure levels. Commit a compact report and require before/after numbers for any optimization.