[shaohaozhi286]

Summary

• aggregate all accounted model-step token usage across one regular agent turn
• expose the same turn summary and turn_id on terminal busy_changed and done session API events
• preserve existing incremental usage and model_step_finish events
• mark mixed provider-reported/estimated summaries with has_data: false while retaining estimated counts and complete context breakdowns
• add OpenSpec proposal/design/spec/tasks plus daemon API documentation

API shape

{
  "type": "done",
  "payload": {
    "outcome": "completed",
    "turn_id": "...",
    "usage": {
      "prompt_tokens": 44100,
      "completion_tokens": 2100,
      "total_tokens": 46200,
      "cache_read_tokens": 32000,
      "cache_write_tokens": 0,
      "reasoning_tokens": 500,
      "has_data": true
    }
  }
}

Terminal usage is a whole-turn summary, not another delta to add to the preceding per-step usage events.

Verification

• npx --yes @fission-ai/openspec@1.12.0 validate expose-turn-token-usage --strict
• bash scripts/code_quality_check.sh
• git diff --check
• added regression coverage for multi-step aggregation and mixed reported/estimated usage

Per request, no C++ compilation/build was run locally; CI should execute the added unit tests.