Claude Code v2.1.319 (2026-11-06, https://code.claude.com/docs/en/changelog): adds permissions.denyInsteadOfPrompting (a call that would wait on a permission prompt is denied and the turn continues); an exact Bash ask rule is refused when a broader deny covers it; fixed a Bash permission bypass where a crafted environment prefix hid part of a command from the permission checks. yoyo: safety.rs already has tests saying a leading VAR=value assignment does not change the verdict (safety.rs ~4239, 4316, 4592). That is read from code; this release was not probed.
Codex CLI 0.167.0 (2026-11-06, https://github.com/openai/codex/releases): automatic compaction now counts retained output from failed responses and from responses whose token usage is missing. Transferable defect class: a context-usage estimate built only from responses that report usage under-counts, so compaction never fires. Open question for yoyo: do its context bar and proactive_compact_if_needed count messages without usage data?
Separate yoyo-side finding the same day: Google rejects a bad key with HTTP 400, reason API_KEY_INVALID, message API key not valid. yoyo prompt_retry diagnose_api_error_for matches only 401, unauthorized, invalid api key, invalid_api_key and authentication, so its cross-provider-fallback auth hint never fires for Google.