Audited against the public tracker on 2026-09-08: issues #1–#16 exist and are open. #12, #14 and #15 are implemented; closure notes below are prepared for the maintainer. The others remain genuine gaps, not v0.1.0 blockers. Complex adapter and test work should not be labeled good first issue.
GitHub #17 and #18 are dependency pull requests, not Action/example issues. The Action and package example are implemented in this release pass; no artificial replacement issues are needed.
Run all twelve tests with a current authenticated CLI; attach unmodified results and redacted evidence; correct any documented flag drift.
Runtime remains unknown. Verify the existing API-key mapping, --approval-mode yolo, MCP transports, JSON extraction and skills using suitable existing access. Account eligibility may require enterprise/Cloud or paid API access, but the adapter only forwards GEMINI_API_KEY. Do not buy access or spend paid API credits for this task. Antigravity CLI is separate. This task includes adapter/auth diagnosis and is not a good first issue.
Confirm native read configuration and headless writing with an authenticated provider; document any mapping differences.
Investigate supported executable automation, implement only practical interaction mappings, and leave GUI-only cases unknown.
Determine supported headless integration; provide a documented adapter and first real result.
Verify current official invocation and authentication, implement contract, and run nested instructions.
Document current practical automation options; keep manual documentation separate from observed harness results.
Run nested, root, precedence, and headless tests; record exact Windows and CLI versions.
Run the twelve priority tests and include complete evidence bundles.
Check subprocess lifecycle, paths, configuration, and structured-output extraction.
Run both transports using a documented authentication method; inspect server receipts.
Design paired deterministic success/failure invocations; distinguish provider errors from usable process semantics.
Prepared closure: execution.exit-success@1.0.0 checks the success marker and zero exit; execution.exit-failure@1.0.0 checks that the failure task was reached before judging non-zero exit. Unreached tasks, timeouts and adapter-reported execution errors are ERROR, not task-semantic PASS. Generic-core regression checks cover these paths. Real Linux runs for Codex, OpenCode and Antigravity retain the observed zero-exit failure behavior. See success, failure and observations. Scope: process exit signaling, not universal recognition of provider failures after the marker; adapters must report native execution errors. Close as completed once this pass is reviewed.
Use a harmless marker-writing lifecycle hook; map native configurations in adapters only.
Create a disposable Git repo and linked worktree; verify operation without implying security guarantees.
Prepared closure: worktree setup/evaluator create a disposable primary checkout and detached linked worktree. The evaluator requires the worktree token, unchanged primary token and recorded Git root. Generic-core checks and real Linux observations exercise it. This proves only the requested file behavior, not instruction resolution or security isolation. Close as completed.
Use benign bounded file edits and temporary sentinels; describe observed behavior without security claims.
Prepared closure: scope setup/evaluator create a temporary Git repo, nested target, and parent/sibling sentinels. The evaluator requires the target token and preserved sentinel contents; regression checks and real Linux observations exist. This observes selected files, not every possible write or a security boundary. Close as completed.
Use Job Objects or an equally contained mechanism; verify descendants terminate after a timeout without paid credentials.