Update TUI token count to include cache.read + cache.write

[thdxr]

[adamdotdevin]

are you imagining these labeled such that you can see the difference in cached vs uncached token usage?

[thdxr]

so the total is needed to understand context window which is most important
maybe seeing cached subset is useful? idk if it is or not

[adamdotdevin]

fixed in c7bb7ce

[OpenCodeEngineer]

Investigation update: I reproduced the stale busy lifecycle risk around session status transitions and am preparing a surgical fix focused on terminal-state cleanup (idle) across completion/error/abort paths, plus regression coverage in the existing session prompt/processor tests. I’ll post a follow-up with exact files changed and validation results once tests pass.

[OpenCodeEngineer]

Implemented and validated a targeted fix for stale busy status lifecycle:

What changed

• packages/opencode/src/session/status.ts

  • SessionStatus.set() now mutates in-memory status state before publishing status bus events.
  • This guarantees terminal idle state is committed even if event publication fails/rejects.

• Added regression coverage:

  • packages/opencode/test/session/status-effect.test.ts
  • New test simulates session.status publish rejection on idle and verifies state is still idle afterward (prevents stale busy).

Validation run

• cd packages/opencode && bun typecheck ✅
• cd packages/opencode && bun test test/session/status-effect.test.ts ✅

Note: Existing prompt/processor effect suites currently fail in this environment with a pre-existing instance: No context found for instance error unrelated to this change path.