Codexometer v0.9.0
Highlights
- Add optional DigBench support, discovering all available games from
/gamesat launch and hiding the suite whenDIGBENCH_API_TOKENis absent. - Run selected DigBench games across scoped model/reasoning combinations with exact remote-session confirmation, live level/step progress, authoritative win detection, and stoppable runs.
- Organize built-in benchmarks into Codexometer Core and Codexometer Extended, with per-task scope selection plus clearer Run Scope and Run All behavior.
- Expand benchmark run details with a navigable, copyable workflow transcript covering safe prompts, tool requests/responses, moves, states, final responses, usage, and API-equivalent cost.
- Support isolated benchmark API-key authentication in preference to the prevailing Codex login when supplied, while keeping DigBench credentials out of spawned Codex environments.
- Harden privacy and reliability with benchmark-only bounded capture, runtime-identifier and terminal-control redaction, whole-run timeouts, safe interruption, and resilient remote-session handling.
See PR #21 for the complete implementation and validation details.
Full changelog: v0.8.0...v0.9.0