Dashboard › opencode-lore › Distillation
23b5dba2-9bee-4805-a44c-5d7ab440c530["lore_tm_v1_YKyZqPU4Ya3XZYOxO6RnmP7V2fejabuqFUF2ObUkQjI","lore_tm_v1_q7A3m5Fapx1W4MDqbWbZ2hphOqnSRDhIAVxu54bVkcU","lore_tm_v1_FhUlD42Nk-7VFzh8EOPH3xnBA_-Y5AJBtCz5jbJD6vg","lore_tm_v1_Pgnk6BlsOsNwzjy1mpB-JkQdbN_Bs_SWedleLzdBkDc","lore_tm_v1__Qu33ngwgnvbmfJdYKu0HtXBMWhgGE1mLvGxvetQTKg"]
Date: Sep 16, 2026
packages/gateway/test/llm-adapter.test.ts enforces this in empty worker diagnostics never include response values, where the sentinel is absent and finish_reason=unknown is present, alongside tests ensuring malformed JSON and worker diagnostics do not expose response content, provider bodies, statusText, URL credentials/query, body-controlled fields, or content-type values.packages/gateway/test/claude-usage-headers.test.ts test never reuses another account's quota when an upstream omits it.packages/gateway/test/openai-responses-stream.test.ts test uses created metadata and failure usage in the plain streaming route verifies that a public response.failed keeps trusted metadata from response.created (id: "resp_plain_trusted_failure", created_at: 123, model: "gpt-5.6-sol"), retains terminal usage as { inputTokens: 5, outputTokens: 2 }, produces empty content and rawOutputItems, reports successful: false, excludes private terminal ID/model/diagnostic values from output, completion data, and logs, and logs exactly "openai-responses provider failure".packages/gateway/test/openai-responses-stream.test.ts test projects trusted in-progress metadata in the plain streaming route verifies that hostile model "private-plain-in-progress-model" from response.in_progress is excluded from streamed output and completion accounting; the failed completion instead retains trusted created-state id: "resp_plain_in_progress_failure" and model: "gpt-5.6-sol", empty content/rawOutputItems, usage { inputTokens: 5, outputTokens: 2 }, and successful: false.packages/gateway/test/openai-responses-stream.test.ts verifies Codex incomplete-response mappings: a response.incomplete with usage { input_tokens: 40, output_tokens: 100 } maps to stopReason: "max_tokens" and outputTokens: 100; strict empty incomplete responses under validation: "public" and stopAtTerminal: true require no output items, map incomplete_details.reason: "max_output_tokens" to "max_tokens" and "content_filter" to "content_filter", and return content: [].