Distillation

Date: Sep 16, 2026

• 🟡 (15:42) Candidate diff output showed extensive regression-test additions around streamResponsesPassthrough, including hostile provider-failure payloads, response.created and response.in_progress SSE events, terminal-frame handling, completion callbacks collecting { response, successful }, sequence-number validation, and sanitization of nested error: { type: "provider_error", message: secret } data.
• 🟡 (15:42) One visible passthrough test used hostileModel = "private-plain-in-progress-model" while the trusted response metadata used model "gpt-5.6-sol", response ID "resp_plain_in_progress_failure", created_at: 123, and status "in_progress".
• 🟡 (15:42) Additional visible test fixtures included response ID "resp_provider_failure", sequence_number: 99, output_index: 0, content_index: 0, and usage { input_tokens: 11, output_tokens: 4 }.
• 🟡 (15:42) Candidate tests included cleanup via afterEach(() => setRecallContinuationFailureHook(undefined)), indicating explicit reset of the recall-continuation failure hook between tests.
• 🟡 (15:42) The complete truncated tool output was saved at /home/byk/.local/share/opencode/tool-output/tool_0aae205b1001iPKkRuTeH6Y2w4.