harness-compat.json

{
  "total": 1225,
  "passed": 427,
  "needThink": 991,
  "needThinkPassed": 193,
  "files": {
    "action-pause-recovery.test.ts": {
      "total": 3,
      "passed": 0
    },
    "actions-attach-reply.test.ts": {
      "total": 12,
      "passed": 0
    },
    "actions-durable-pause.test.ts": {
      "total": 40,
      "passed": 2
    },
    "agent-tool-reattach-recovery.test.ts": {
      "total": 2,
      "passed": 0
    },
    "agent-tool-rebind-noop.test.ts": {
      "total": 6,
      "passed": 0
    },
    "agent-tools.test.ts": {
      "total": 47,
      "passed": 12
    },
    "assistant-agent-loop.test.ts": {
      "total": 26,
      "passed": 8
    },
    "assistant-agent.test.ts": {
      "total": 5,
      "passed": 5
    },
    "assistant-message-metadata.test.ts": {
      "total": 9,
      "passed": 1
    },
    "assistant-tools.test.ts": {
      "total": 34,
      "passed": 34
    },
    "attachment-consumption.test.ts": {
      "total": 4,
      "passed": 0
    },
    "browser-tools.test.ts": {
      "total": 6,
      "passed": 6
    },
    "channel-policy.test.ts": {
      "total": 3,
      "passed": 2
    },
    "channel-recovery.test.ts": {
      "total": 5,
      "passed": 0
    },
    "channel-threading.test.ts": {
      "total": 4,
      "passed": 1
    },
    "channels.test.ts": {
      "total": 8,
      "passed": 8
    },
    "client-tools.test.ts": {
      "total": 100,
      "passed": 31
    },
    "cold-rpc-initialization.test.ts": {
      "total": 4,
      "passed": 3
    },
    "connection-state.test.ts": {
      "total": 2,
      "passed": 2
    },
    "deliver-notice.test.ts": {
      "total": 7,
      "passed": 0
    },
    "errored-stream-replay.test.ts": {
      "total": 1,
      "passed": 0
    },
    "eviction-recovery.test.ts": {
      "total": 1,
      "passed": 0
    },
    "execute-hitl.test.ts": {
      "total": 12,
      "passed": 0
    },
    "execute-tool.test.ts": {
      "total": 6,
      "passed": 3
    },
    "extension-manager.test.ts": {
      "total": 34,
      "passed": 34
    },
    "fetch-tools.test.ts": {
      "total": 51,
      "passed": 50
    },
    "fiber.test.ts": {
      "total": 7,
      "passed": 7
    },
    "hooks.test.ts": {
      "total": 140,
      "passed": 28
    },
    "hydration-budget.test.ts": {
      "total": 9,
      "passed": 1
    },
    "max-concurrent-agent-tools.test.ts": {
      "total": 5,
      "passed": 1
    },
    "mcp-tool-materialization.test.ts": {
      "total": 2,
      "passed": 0
    },
    "media-eviction.test.ts": {
      "total": 16,
      "passed": 3
    },
    "message-reconciliation.test.ts": {
      "total": 11,
      "passed": 4
    },
    "messengers.test.ts": {
      "total": 103,
      "passed": 79
    },
    "nested-agent-tools.test.ts": {
      "total": 3,
      "passed": 0
    },
    "onconnect-broadcast.test.ts": {
      "total": 13,
      "passed": 1
    },
    "onstart-degraded.test.ts": {
      "total": 5,
      "passed": 1
    },
    "origin-message-ids.test.ts": {
      "total": 8,
      "passed": 0
    },
    "prompt-cache.test.ts": {
      "total": 10,
      "passed": 3
    },
    "run-turn-recovery.test.ts": {
      "total": 10,
      "passed": 1
    },
    "run-turn.test.ts": {
      "total": 34,
      "passed": 13
    },
    "scheduled-tasks.test.ts": {
      "total": 17,
      "passed": 1
    },
    "stream-cleanup.test.ts": {
      "total": 5,
      "passed": 0
    },
    "streaming-message-id.test.ts": {
      "total": 1,
      "passed": 1
    },
    "submissions.test.ts": {
      "total": 92,
      "passed": 8
    },
    "telegram-messenger.test.ts": {
      "total": 2,
      "passed": 2
    },
    "terminal-frame-order.test.ts": {
      "total": 3,
      "passed": 0
    },
    "think-eviction.test.ts": {
      "total": 4,
      "passed": 1
    },
    "think-session.test.ts": {
      "total": 276,
      "passed": 65
    },
    "turn-metadata.test.ts": {
      "total": 4,
      "passed": 1
    },
    "workflows.test.ts": {
      "total": 3,
      "passed": 3
    },
    "ws-web-channel.test.ts": {
      "total": 10,
      "passed": 1
    }
  },
  "passing": [
    "actions-durable-pause.test.ts > action() durable-pause validation > allows kind durable-pause without an approval policy",
    "actions-durable-pause.test.ts > action() durable-pause validation > rejects kind durable-pause with approval: false",
    "agent-tools.test.ts > Think agent tools > an Infinity no-progress budget never seals on silence — only the hard ceiling ends the wait (#1630/#1672)",
    "agent-tools.test.ts > Think agent tools > bounds Think recovery when child facet startup never completes",
    "agent-tools.test.ts > Think agent tools > defaults the re-attach hard ceiling to uncapped (Infinity) when unset (#1630/#1672)",
    "agent-tools.test.ts > Think agent tools > does not re-arm after a full no-progress window even if the child progressed earlier (#1630)",
    "agent-tools.test.ts > Think agent tools > eventDelivery: \"terminal\" (#2298) > rejects terminal delivery for a detached run",
    "agent-tools.test.ts > Think agent tools > honors the public AgentStaticOptions re-attach budgets (#1630)",
    "agent-tools.test.ts > Think agent tools > keeps scheduled Think startup recovery single-flight",
    "agent-tools.test.ts > Think agent tools > lets Think startup return before stale child recovery finalizes",
    "agent-tools.test.ts > Think agent tools > only recovers rows that were stale before Think startup began",
    "agent-tools.test.ts > Think agent tools > re-arms across a clean mid-flight stream-close and follows an advancing child to completed (#1630)",
    "agent-tools.test.ts > Think agent tools > re-attach returns not-tailable for an adapter without a live-tail (#1630)",
    "agent-tools.test.ts > Think agent tools > runs scheduled Think startup recovery and finalizes stale rows",
    "assistant-agent-loop.test.ts > Think — agentic loop > context assembly > converts messages to model format",
    "assistant-agent-loop.test.ts > Think — agentic loop > default loop — text only > persists assistant message after streaming",
    "assistant-agent-loop.test.ts > Think — agentic loop > default loop — text only > streams a response using the mock model",
    "assistant-agent-loop.test.ts > Think — agentic loop > default loop — with tools > custom maxSteps property is respected",
    "assistant-agent-loop.test.ts > Think — agentic loop > default loop — with tools > executes a tool and returns text after",
    "assistant-agent-loop.test.ts > Think — agentic loop > defaultContextOverflowClassifier > classifies common provider context-overflow errors",
    "assistant-agent-loop.test.ts > Think — agentic loop > defaultContextOverflowClassifier > returns undefined for unrelated errors",
    "assistant-agent-loop.test.ts > Think — agentic loop > getModel() error > returns an error when getModel is not overridden",
    "assistant-agent.test.ts > Think — cancel > cancel message does not crash the agent",
    "assistant-agent.test.ts > Think — clear > clears all messages",
    "assistant-agent.test.ts > Think — message persistence > messages survive across agent instances",
    "assistant-agent.test.ts > Think — streaming flow > persists assistant message after streaming",
    "assistant-agent.test.ts > Think — streaming flow > sends a chat request and receives streamed response",
    "assistant-message-metadata.test.ts > Think — server-authored assistant-message metadata > stamps no metadata keys when no writer is configured (opt-in)",
    "assistant-tools.test.ts > assistant tools — bash > does not overwrite workspace files skipped during mounting",
    "assistant-tools.test.ts > assistant tools — bash > does not persist synthetic shell paths (/bin, /usr, …) to the workspace",
    "assistant-tools.test.ts > assistant tools — bash > mounts workspace directories beyond the first readDir page",
    "assistant-tools.test.ts > assistant tools — bash > persists empty directory creates and deletes",
    "assistant-tools.test.ts > assistant tools — bash > returns non-zero bash exits with output",
    "assistant-tools.test.ts > assistant tools — bash > returns structured output for bash timeouts",
    "assistant-tools.test.ts > assistant tools — bash > runs bash against workspace files and persists changes",
    "assistant-tools.test.ts > assistant tools — edit > creates new file with empty old_string",
    "assistant-tools.test.ts > assistant tools — edit > fuzzy matches on whitespace differences",
    "assistant-tools.test.ts > assistant tools — edit > replaces exact match",
    "assistant-tools.test.ts > assistant tools — edit > returns error for ambiguous fuzzy match",
    "assistant-tools.test.ts > assistant tools — edit > returns error for missing file",
    "assistant-tools.test.ts > assistant tools — edit > returns error when old_string has multiple matches",
    "assistant-tools.test.ts > assistant tools — edit > returns error when old_string not found",
    "assistant-tools.test.ts > assistant tools — find > finds files by glob pattern",
    "assistant-tools.test.ts > assistant tools — grep > returns context lines when requested",
    "assistant-tools.test.ts > assistant tools — grep > searches file contents with regex",
    "assistant-tools.test.ts > assistant tools — grep > searches with fixed string",
    "assistant-tools.test.ts > assistant tools — grep > skips files larger than 1 MB",
    "assistant-tools.test.ts > assistant tools — grep > supports case-sensitive search",
    "assistant-tools.test.ts > assistant tools — list > lists files and directories",
    "assistant-tools.test.ts > assistant tools — read > does not line-number unknown binary files",
    "assistant-tools.test.ts > assistant tools — read > reads a file with line numbers",
    "assistant-tools.test.ts > assistant tools — read > returns compact image metadata and image-data model content",
    "assistant-tools.test.ts > assistant tools — read > returns error for directory",
    "assistant-tools.test.ts > assistant tools — read > returns error for missing file",
    "assistant-tools.test.ts > assistant tools — read > returns file-data model output for PDFs",
    "assistant-tools.test.ts > assistant tools — read > sniffs gif image bytes when MIME is generic",
    "assistant-tools.test.ts > assistant tools — read > sniffs jpeg image bytes when MIME is generic",
    "assistant-tools.test.ts > assistant tools — read > sniffs png image bytes when MIME is generic",
    "assistant-tools.test.ts > assistant tools — read > sniffs webp image bytes when MIME is generic",
    "assistant-tools.test.ts > assistant tools — read > supports offset and limit",
    "assistant-tools.test.ts > assistant tools — write > creates parent directories",
    "assistant-tools.test.ts > assistant tools — write > writes a file and reports stats",
    "browser-tools.test.ts > createBrowserTools > accepts cdpUrl instead of browser binding (Quick Actions skipped without a binding)",
    "browser-tools.test.ts > createBrowserTools > accepts optional timeout and session mode",
    "browser-tools.test.ts > createBrowserTools > exposes the runtime handle and connector via createBrowserRuntime",
    "browser-tools.test.ts > createBrowserTools > omits Quick Action tools when quickActions is false",
    "browser-tools.test.ts > createBrowserTools > requires a browser binding or cdpUrl",
    "browser-tools.test.ts > createBrowserTools > returns browser_execute plus the default Quick Action tools when a binding is present",
    "channel-policy.test.ts > per-channel policy (Phase 4b) > does not apply channel policy for turns on other channels",
    "channel-policy.test.ts > per-channel policy (Phase 4b) > narrows the tool set via the channel policy (removes, not just adds)",
    "channel-threading.test.ts > channel threading (Phase 4a) > has no channel context when none is supplied",
    "channels.test.ts > resolveChannels > absorbs getMessengers() entries as messenger channels and feeds the runtime",
    "channels.test.ts > resolveChannels > allows overriding the web channel's policy with a kind: web entry",
    "channels.test.ts > resolveChannels > always includes an implicit web channel",
    "channels.test.ts > resolveChannels > feeds messenger-kind configureChannels entries into the runtime",
    "channels.test.ts > resolveChannels > registers configureChannels web/voice entries without feeding the runtime",
    "channels.test.ts > resolveChannels > throws on a duplicate id across configureChannels() and getMessengers()",
    "channels.test.ts > resolveChannels > throws when configureChannels() replaces web with a non-web kind",
    "channels.test.ts > resolveChannels > throws when getMessengers() declares a messenger named web",
    "client-tools.test.ts > Think — auto-continuation > autoContinue: true triggers continuation",
    "client-tools.test.ts > Think — auto-continuation > continues after approving a tool that shares its message with a settled tool (#2185)",
    "client-tools.test.ts > Think — auto-continuation > does not continue a standalone errored tool (autoContinue:false, no opted-in sibling) (#1650)",
    "client-tools.test.ts > Think — auto-continuation > does not execute a rejected server approval tool and retains denial",
    "client-tools.test.ts > Think — auto-continuation > persists an approved server tool execution error as terminal output",
    "client-tools.test.ts > Think — auto-continuation > sends a failed continuation's error frame before onChatResponse returns (#2381)",
    "client-tools.test.ts > Think — auto-continuation > treats a pending approved tool as complete — no spurious transcript-repair backstop (#1627)",
    "client-tools.test.ts > Think — auto-continuation > without autoContinue, no continuation",
    "client-tools.test.ts > Think — custom body via WebSocket > body fields persist and are available after turn",
    "client-tools.test.ts > Think — custom body via WebSocket > body is cleared when request has no custom fields",
    "client-tools.test.ts > Think — messageConcurrency > only applies to submit-message, not regenerate-message",
    "client-tools.test.ts > Think — messageConcurrency > post-clear submits are not treated as overlapping",
    "client-tools.test.ts > Think — messageConcurrency > queue: processes all overlapping submits in order",
    "client-tools.test.ts > Think — onChatResponse via WebSocket > fires onChatResponse after WebSocket chat request",
    "client-tools.test.ts > Think — regeneration > multiple regenerations create multiple branches",
    "client-tools.test.ts > Think — regeneration > regenerate-message creates a sibling branch, not a replacement",
    "client-tools.test.ts > Think — regeneration > regenerating an earlier response leaves later stored answers untouched (#2028)",
    "client-tools.test.ts > Think — regeneration > regeneration fires onChatResponse",
    "client-tools.test.ts > Think — regeneration > regeneration preserves conversation history before the branch point",
    "client-tools.test.ts > Think — tool approval > does NOT update tool in output-available state",
    "client-tools.test.ts > Think — tool approval > non-existent toolCallId is a no-op",
    "client-tools.test.ts > Think — tool approval > preserves approval data",
    "client-tools.test.ts > Think — tool broadcast and persistence > other tabs receive continuation stream chunks",
    "client-tools.test.ts > Think — tool broadcast and persistence > tool state survives across agent instances",
    "client-tools.test.ts > Think — tool result application > applies to tool in approval-responded state",
    "client-tools.test.ts > Think — tool result application > does NOT update tool in output-available state",
    "client-tools.test.ts > Think — tool result application > does NOT update tool in output-denied state",
    "client-tools.test.ts > Think — tool result application > sets output-error state with errorText",
    "client-tools.test.ts > Think — tool result application > updates tool part to output-available",
    "client-tools.test.ts > resume coordination during pending continuation > correlates idle STREAM_RESUME_NONE after a continuation completes",
    "client-tools.test.ts > resume coordination during pending continuation > does not send STREAM_RESUME_NONE for an immediate resume request after tool output",
    "cold-rpc-initialization.test.ts > native RPC initialization > hydrates persisted Think history when native RPC is first after eviction",
    "cold-rpc-initialization.test.ts > native RPC initialization > initializes Think before the first inherited RPC method executes",
    "cold-rpc-initialization.test.ts > native RPC initialization > initializes Think before the first subclass RPC method executes",
    "connection-state.test.ts > preserves onConnect state for the first Think root client message",
    "connection-state.test.ts > preserves onConnect state for the first Think sub-agent client message",
    "execute-tool.test.ts > execute tool on the codemode runtime > maps needsApproval AI SDK tools to requiresApproval — the call pauses durably",
    "execute-tool.test.ts > execute tool on the codemode runtime > runs sandbox code against tools.* (ToolSetConnector)",
    "execute-tool.test.ts > execute tool on the codemode runtime > surfaces sandbox errors as error outcomes with an executionId",
    "extension-manager.test.ts > ExtensionManager > getTools > should merge tools from multiple extensions",
    "extension-manager.test.ts > ExtensionManager > getTools > should prefix tool names with extension name",
    "extension-manager.test.ts > ExtensionManager > getTools > should return AI SDK tools from loaded extensions",
    "extension-manager.test.ts > ExtensionManager > getTools > should return empty object when no extensions loaded",
    "extension-manager.test.ts > ExtensionManager > lifecycle after unload > should allow reloading after unload",
    "extension-manager.test.ts > ExtensionManager > lifecycle after unload > should not include tools from unloaded extensions",
    "extension-manager.test.ts > ExtensionManager > lifecycle after unload > should throw when executing a tool from an unloaded extension",
    "extension-manager.test.ts > ExtensionManager > list > should list all loaded extensions",
    "extension-manager.test.ts > ExtensionManager > list > should return empty array when no extensions loaded",
    "extension-manager.test.ts > ExtensionManager > load and discover > should load a multi-tool extension",
    "extension-manager.test.ts > ExtensionManager > load and discover > should load an extension and discover its tools",
    "extension-manager.test.ts > ExtensionManager > load and discover > should reject duplicate extension names",
    "extension-manager.test.ts > ExtensionManager > name sanitization > sanitizeName should replace non-alphanumeric chars",
    "extension-manager.test.ts > ExtensionManager > name sanitization > sanitizeName should strip leading and trailing underscores",
    "extension-manager.test.ts > ExtensionManager > name sanitization > sanitizeName should throw for empty or whitespace-only names",
    "extension-manager.test.ts > ExtensionManager > name sanitization > should sanitize hyphens in extension names for tool names",
    "extension-manager.test.ts > ExtensionManager > network isolation > should block network by default (no network permission)",
    "extension-manager.test.ts > ExtensionManager > persistence and restore > restore should be idempotent",
    "extension-manager.test.ts > ExtensionManager > persistence and restore > should remove from storage on unload",
    "extension-manager.test.ts > ExtensionManager > persistence and restore > should restore extensions from storage after re-creation",
    "extension-manager.test.ts > ExtensionManager > structured source format > hook receives snapshot data",
    "extension-manager.test.ts > ExtensionManager > structured source format > rejects flat-format source with clear error",
    "extension-manager.test.ts > ExtensionManager > structured source format > should catch and report hook errors",
    "extension-manager.test.ts > ExtensionManager > structured source format > should discover hooks via manifest() RPC",
    "extension-manager.test.ts > ExtensionManager > structured source format > should execute tools from structured format",
    "extension-manager.test.ts > ExtensionManager > structured source format > should invoke hooks via entrypoint.hook()",
    "extension-manager.test.ts > ExtensionManager > structured source format > should load extensions with { tools, hooks } format",
    "extension-manager.test.ts > ExtensionManager > structured source format > should return skipped for unsubscribed hooks",
    "extension-manager.test.ts > ExtensionManager > structured source format > tools-only extension has no hooks",
    "extension-manager.test.ts > ExtensionManager > tool execution > should execute a simple tool",
    "extension-manager.test.ts > ExtensionManager > tool execution > should execute math tools correctly",
    "extension-manager.test.ts > ExtensionManager > tool execution > should propagate tool execution errors",
    "extension-manager.test.ts > ExtensionManager > unload > should return false for unknown extension",
    "extension-manager.test.ts > ExtensionManager > unload > should unload an extension",
    "fetch-tools.test.ts > Think — fetchTools integration > does not register fetch tools by default",
    "fetch-tools.test.ts > createFetchTools — config > generates one fetch_<name> tool per binding with a sanitized name",
    "fetch-tools.test.ts > createFetchTools — config > registers fetch_url only when a public allowlist is set",
    "fetch-tools.test.ts > createFetchTools — config > throws when neither allowlist nor bindings are configured",
    "fetch-tools.test.ts > fetch — abort and timeout > reports aborted when the turn signal aborts",
    "fetch-tools.test.ts > fetch — abort and timeout > times out a slow request",
    "fetch-tools.test.ts > fetch — header allowlist > forwards allowlisted model headers and fixed headers, drops the rest",
    "fetch-tools.test.ts > fetch — header allowlist > lets fixed binding headers take precedence over model headers",
    "fetch-tools.test.ts > fetch — header allowlist > lets the model override the default Accept header",
    "fetch-tools.test.ts > fetch — header allowlist > sends a markdown-first default Accept header when none is set",
    "fetch-tools.test.ts > fetch — header allowlist > sends no default Accept header when disabled",
    "fetch-tools.test.ts > fetch — observability > emits an event on success and on block",
    "fetch-tools.test.ts > fetch — redirects > blocks cross-origin redirects for bindings",
    "fetch-tools.test.ts > fetch — redirects > does not follow redirects when policy is none",
    "fetch-tools.test.ts > fetch — redirects > follows same-origin redirects through the binding",
    "fetch-tools.test.ts > fetch — size limits > returns too_large for JSON that exceeds the byte cap",
    "fetch-tools.test.ts > fetch — size limits > truncates oversized text and flags truncated",
    "fetch-tools.test.ts > fetch — workspace spill > writes binary bodies via writeFileBytes when spill is enabled in auto",
    "fetch-tools.test.ts > fetch — workspace spill > writes text bodies to the workspace in workspace mode",
    "fetch-tools.test.ts > fetch_<binding> — responses > parses JSON bodies",
    "fetch-tools.test.ts > fetch_<binding> — responses > rejects binary content in auto mode without spill",
    "fetch-tools.test.ts > fetch_<binding> — responses > rejects paths outside the binding allowlist",
    "fetch-tools.test.ts > fetch_<binding> — responses > reports invalid JSON",
    "fetch-tools.test.ts > fetch_<binding> — responses > returns HTML as text in auto mode",
    "fetch-tools.test.ts > fetch_<binding> — responses > returns empty text for an empty (204) response",
    "fetch-tools.test.ts > fetch_<binding> — responses > returns text bodies",
    "fetch-tools.test.ts > fetch_<binding> — responses > surfaces non-2xx responses",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target http://0.0.0.0/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target http://0177.0.0.1/ even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target http://0x7f.0.0.1/ even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target http://10.0.0.1/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target http://100.64.0.1/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target http://127.0.0.1/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target http://127.1/ even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target http://169.254.169.254/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target http://172.16.0.1/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target http://192.168.1.1/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target http://2130706433/ even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target https://[::1]/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target https://[::ffff:127.0.0.1]/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target https://[::ffff:192.168.0.1]/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target https://[fc00::1]/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target https://[fe80::1]/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target https://app.localhost/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target https://foo.internal/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > blocks private/local target https://localhost/x even when allowlisted",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > normalizes a bare-origin allowlist entry to match the origin and subpaths",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > rejects URLs that are not on the allowlist",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > rejects URLs with embedded credentials",
    "fetch-tools.test.ts > fetch_url — allowlist + SSRF > rejects invalid URLs",
    "fiber.test.ts > Think — fiber recovery > should clean up the fiber row after recovery",
    "fiber.test.ts > Think — fiber recovery > should detect and recover an interrupted fiber",
    "fiber.test.ts > Think — runFiber > should clean up the fiber row on completion",
    "fiber.test.ts > Think — runFiber > should propagate errors from fiber execution",
    "fiber.test.ts > Think — runFiber > should run a fiber to completion and return the result",
    "fiber.test.ts > Think — runFiber > should support checkpointing via ctx.stash()",
    "fiber.test.ts > Think — runFiber > should support fire-and-forget fibers",
    "hooks.test.ts > Think — ToolCallDecision honored by wrapped execute > 'substitute' short-circuits execute and returns the substituted output",
    "hooks.test.ts > Think — ToolCallDecision honored by wrapped execute > async beforeToolCall (Promise<ToolCallDecision>) is awaited correctly",
    "hooks.test.ts > Think — ToolCallDecision honored by wrapped execute > collapses Promise<AsyncIterable> returns to the last yielded value",
    "hooks.test.ts > Think — ToolCallDecision honored by wrapped execute > collapses sync AsyncIterable returns to the last yielded value",
    "hooks.test.ts > Think — ToolCallDecision honored by wrapped execute > void decision runs the original execute with original input",
    "hooks.test.ts > Think — beforeToolCall gates whether execute runs > never runs execute when the result is substituted",
    "hooks.test.ts > Think — beforeToolCall gates whether execute runs > runs execute exactly once on a void decision",
    "hooks.test.ts > Think — beforeToolCall gates whether execute runs > runs execute exactly once when 'allow' substitutes the input",
    "hooks.test.ts > Think — beforeTurn config overrides > allows TurnConfig to send reasoning when the instance default is false",
    "hooks.test.ts > Think — beforeTurn config overrides > errors a wait-mode continuation whose structured output does not parse",
    "hooks.test.ts > Think — beforeTurn config overrides > errors a wait-mode turn whose structured output does not parse (#2263)",
    "hooks.test.ts > Think — beforeTurn config overrides > sends reasoning chunks by default on the chat() path",
    "hooks.test.ts > Think — beforeTurn hook > captures continuation flag from programmatic path",
    "hooks.test.ts > Think — beforeTurn hook > fires from chat() sub-agent path",
    "hooks.test.ts > Think — beforeTurn hook > fires on every turn",
    "hooks.test.ts > Think — beforeTurn hook > keeps the default gateway when getGateway returns undefined",
    "hooks.test.ts > Think — beforeTurn hook > returns a LanguageModel object unchanged",
    "hooks.test.ts > Think — beforeTurn hook > routes @hf/ ids to Workers AI like @cf/ ids",
    "hooks.test.ts > Think — dynamic context > dynamic block visible in chat turn tools",
    "hooks.test.ts > Think — hook convergence > beforeTurn fires from chat() RPC path",
    "hooks.test.ts > Think — hook convergence > beforeTurn fires from saveMessages path",
    "hooks.test.ts > Think — maxSteps property > respects maxSteps override on class",
    "hooks.test.ts > Think — maxSteps property > works with tool-calling loop agent",
    "hooks.test.ts > Think — onChunk hook > fires for streaming chunks",
    "hooks.test.ts > Think — raw needsApproval tool + beforeToolCall ordering > runs execute exactly once when both gates allow",
    "hooks.test.ts > Think — streaming tool preservation > collapses a Promise<AsyncIterable> (non-canonical) execute",
    "hooks.test.ts > Think — tool-call hooks expose typed input/output > beforeToolCall and tool execute see the turn's request id",
    "hooks.test.ts > Think — tool-call hooks expose typed input/output > beforeToolCall receives toolName and typed input",
    "hydration-budget.test.ts > hydrationByteBudget — windowed hydration (#1710) > chat works on a windowed-boot agent and persists past the window",
    "max-concurrent-agent-tools.test.ts > maxConcurrentAgentTools > DOES count an in-flight `running` run toward the cap",
    "media-eviction.test.ts > hasEvictableMedia — the in-memory gate for scheduling a pass > ignores text, markers, remote urls and small payloads",
    "media-eviction.test.ts > hasEvictableMedia — the in-memory gate for scheduling a pass > sees a payload nested inside a tool output",
    "media-eviction.test.ts > hasEvictableMedia — the in-memory gate for scheduling a pass > sees an inline file payload at or above the threshold",
    "message-reconciliation.test.ts > Think — message reconciliation on incoming submits > collapses an optimistic copy of a provider-executed tool with a large input",
    "message-reconciliation.test.ts > Think — message reconciliation on incoming submits > does not re-repair an already-errored tool call or clobber its errorText",
    "message-reconciliation.test.ts > Think — message reconciliation on incoming submits > merges server tool outputs into a stale client snapshot when IDs match",
    "message-reconciliation.test.ts > Think — message reconciliation on incoming submits > preserves a denied tool approval (output-denied) instead of flipping it to errored",
    "messengers.test.ts > telegram messenger provider > classifies Telegram no-op edit errors",
    "messengers.test.ts > telegram messenger provider > requires explicit webhook verification posture",
    "messengers.test.ts > telegram messenger provider > splits long Telegram follow-up text without dropping content",
    "messengers.test.ts > telegram messenger provider > supports distinct Telegram adapter names and rejects duplicate defaults",
    "messengers.test.ts > telegram messenger provider > verifies Telegram secret token headers",
    "messengers.test.ts > think messengers core > accepts a custom channelSpeakerLabel formatter",
    "messengers.test.ts > think messengers core > attachment fetch > copies a view backed by a SharedArrayBuffer instead of dropping it",
    "messengers.test.ts > think messengers core > attachment fetch > copies only a Buffer view's own bytes out of its backing store",
    "messengers.test.ts > think messengers core > attachment fetch > returns a view's own buffer without copying when the view spans all of it",
    "messengers.test.ts > think messengers core > attachment fetch > returns an ArrayBuffer from the adapter unchanged",
    "messengers.test.ts > think messengers core > attachment fetch > returns an empty ArrayBuffer when the adapter resolves nothing",
    "messengers.test.ts > think messengers core > burst replies end to end (#2312) > finds the private Chat SDK queue methods the recovery drain relies on",
    "messengers.test.ts > think messengers core > carries an additive delivery tag without changing recovery classification",
    "messengers.test.ts > think messengers core > checkpoints completed before posting the terminal reply when failed after text (#1842)",
    "messengers.test.ts > think messengers core > checkpoints completed before posting the terminal reply when failed before text (#1842)",
    "messengers.test.ts > think messengers core > checkpoints completed before posting the terminal reply when interrupted (#1842)",
    "messengers.test.ts > think messengers core > classifies messenger delivery failures",
    "messengers.test.ts > think messengers core > converts messenger actions to Think user messages",
    "messengers.test.ts > think messengers core > converts messenger events to Think user messages with attachments",
    "messengers.test.ts > think messengers core > creates serializable recovery snapshots without live raw data",
    "messengers.test.ts > think messengers core > delivers successful replies with active messenger context and overflow chunks",
    "messengers.test.ts > think messengers core > derives stable conversation and idempotency keys",
    "messengers.test.ts > think messengers core > detects mentions on messages the adapter did not flag",
    "messengers.test.ts > think messengers core > does not leak internal error details into messenger replies",
    "messengers.test.ts > think messengers core > falls back through fullName || userName || userId for the default label",
    "messengers.test.ts > think messengers core > gives queued messages a TTL that outlasts a slow turn",
    "messengers.test.ts > think messengers core > honors custom webhook verifier responses before Chat SDK handling",
    "messengers.test.ts > think messengers core > inline attachment data > fetches Blob data through arrayBuffer()",
    "messengers.test.ts > think messengers core > inline attachment data > maps ArrayBuffer and Uint8Array data",
    "messengers.test.ts > think messengers core > inline attachment data > maps Buffer data and fetches it when the adapter has no fetchData",
    "messengers.test.ts > think messengers core > inline attachment data > prefers the adapter's fetchData over inline data",
    "messengers.test.ts > think messengers core > is a no-op when the adapter exposes no botUserId",
    "messengers.test.ts > think messengers core > keeps raw payloads and attachment bytes out of persisted message metadata",
    "messengers.test.ts > think messengers core > leaves attachment id undefined when fetchMetadata has no known id key",
    "messengers.test.ts > think messengers core > leaves other users' resolved mentions untouched",
    "messengers.test.ts > think messengers core > lets custom webhook verification read the body without consuming adapter input",
    "messengers.test.ts > think messengers core > marks messenger reply fibers as streaming only when visible text starts",
    "messengers.test.ts > think messengers core > messages folded by the concurrency strategy (#2312) > carries skipped messages on the event, oldest first, resolving self-mentions in each",
    "messengers.test.ts > think messengers core > messages folded by the concurrency strategy (#2312) > drops an attachment-only run's empty speaker line",
    "messengers.test.ts > think messengers core > messages folded by the concurrency strategy (#2312) > keeps each speaker's own label when a burst mixes senders",
    "messengers.test.ts > think messengers core > messages folded by the concurrency strategy (#2312) > keeps the answered message's id so idempotency is unchanged",
    "messengers.test.ts > think messengers core > messages folded by the concurrency strategy (#2312) > leaves `skipped` off the event when nothing was folded",
    "messengers.test.ts > think messengers core > messages folded by the concurrency strategy (#2312) > lists attachments from skipped messages before the answered message's own",
    "messengers.test.ts > think messengers core > messages folded by the concurrency strategy (#2312) > renders a direct-message burst without speaker labels",
    "messengers.test.ts > think messengers core > messages folded by the concurrency strategy (#2312) > renders one sender's burst under a single speaker label",
    "messengers.test.ts > think messengers core > messages folded by the concurrency strategy (#2312) > serializes skipped messages without live-only fields",
    "messengers.test.ts > think messengers core > never prefixes direct message actions",
    "messengers.test.ts > think messengers core > never prefixes direct messages regardless of channelSpeakerLabel",
    "messengers.test.ts > think messengers core > normalizes inferred defaults",
    "messengers.test.ts > think messengers core > parses and classifies messenger recovery snapshots",
    "messengers.test.ts > think messengers core > posts only the apology, never an empty stream, when an interrupted turn has no text",
    "messengers.test.ts > think messengers core > posts text past the visible limit before the target delivers the rest (#2106)",
    "messengers.test.ts > think messengers core > prefixes channel actions with the resolved speaker label",
    "messengers.test.ts > think messengers core > prefixes channel messages with the default fullName cascade",
    "messengers.test.ts > think messengers core > preserves attachment fetchMetadata and backfills id when converting",
    "messengers.test.ts > think messengers core > preserves delivery errors when cancelling local self targets",
    "messengers.test.ts > think messengers core > recovers interrupted messenger reply fibers through the shared Chat runtime",
    "messengers.test.ts > think messengers core > rejects duplicate adapter names before creating a shared Chat runtime",
    "messengers.test.ts > think messengers core > rejects invalid and duplicate paths",
    "messengers.test.ts > think messengers core > rejects webhook requests when custom verification returns false",
    "messengers.test.ts > think messengers core > reposts the partial when the stream post rejected before the target delivers the rest (#2106)",
    "messengers.test.ts > think messengers core > requires an explicit webhook verification posture",
    "messengers.test.ts > think messengers core > resolves the self-mention in default events using the bot handle",
    "messengers.test.ts > think messengers core > rewrites the bot's own self-mention to the bot handle",
    "messengers.test.ts > think messengers core > separates text segments across tool-call boundaries (#1841)",
    "messengers.test.ts > think messengers core > skips the apology when the target delivers the recovered reply itself (#2106)",
    "messengers.test.ts > think messengers core > streamed replies on adapters without native streaming > keeps its own adapter when another runtime registers its Chat mid-recovery",
    "messengers.test.ts > think messengers core > streamed replies on adapters without native streaming > leaves native streaming untouched",
    "messengers.test.ts > think messengers core > streamed replies on adapters without native streaming > posts real reply text first instead of a `...` placeholder (#2310)",
    "messengers.test.ts > think messengers core > streamed replies on adapters without native streaming > posts the empty-response text, not a blank message, when a turn produces no text",
    "messengers.test.ts > think messengers core > streamed replies on adapters without native streaming > recovers a reply through its own adapter after another runtime registered its Chat",
    "messengers.test.ts > think messengers core > streams visible text while retaining overflow",
    "messengers.test.ts > think messengers core > surfaces the interrupted apology, not a truncated final reply, when the model turn is interrupted by recovery (#1644)",
    "messengers.test.ts > think messengers core > typing indicator (#2324) > never overlaps slow typing requests or lets one land after the first post",
    "messengers.test.ts > think messengers core > typing indicator (#2324) > refreshes until the first text, then stops",
    "messengers.test.ts > think messengers core > typing indicator (#2324) > still posts the reply when the first typing request never settles",
    "messengers.test.ts > think messengers core > typing indicator (#2324) > still posts the reply when the refresh typing request never settles",
    "messengers.test.ts > think messengers core > typing indicator (#2324) > still runs the turn when the typing indicator fails",
    "messengers.test.ts > think messengers core > typing indicator (#2324) > stops refreshing when the turn fails",
    "onconnect-broadcast.test.ts > Think — onConnect broadcast policy > broadcasts CHAT_MESSAGES on connect when no stream is active",
    "onstart-degraded.test.ts > Think onStart degradation (#1710) > scheduled-task reconcile failure > chat still works on the degraded agent",
    "prompt-cache.test.ts > prompt-cache prefix stability (#2200) > extends the previous request every turn with no reduction",
    "prompt-cache.test.ts > prompt-cache prefix stability (#2200) > never rewrites the prefix when truncationStep is Infinity",
    "prompt-cache.test.ts > prompt-cache prefix stability (#2200) > rewrites the prefix once per compaction",
    "run-turn-recovery.test.ts > recovery × runTurn > wraps a runTurn turn in a recovery fiber and cleans it up",
    "run-turn.test.ts > Think — runTurn > concurrent non-nested wait enqueues behind an active turn",
    "run-turn.test.ts > Think — runTurn > emits no chat:turn metadata for skipped empty wait input",
    "run-turn.test.ts > Think — runTurn > omits a persisted partial message from an aborted turn result",
    "run-turn.test.ts > Think — runTurn > returns each queued wait turn's own assistant message",
    "run-turn.test.ts > Think — runTurn > returns persisted content when the response hook mutates history",
    "run-turn.test.ts > Think — runTurn > returns stored tool output with toJSON without suppressing the response hook",
    "run-turn.test.ts > Think — runTurn > stream mode drives callback and resolves void like chat",
    "run-turn.test.ts > Think — runTurn > stream mode supports array input",
    "run-turn.test.ts > Think — runTurn > stream mode supports function input evaluated inside the turn",
    "run-turn.test.ts > Think — runTurn > validates submit and stream require input",
    "run-turn.test.ts > Think — runTurn > wait mode returns TurnResult equivalent to saveMessages",
    "run-turn.test.ts > Think — runTurn > wait mode supports array input",
    "run-turn.test.ts > Think — runTurn > wait mode supports function input",
    "scheduled-tasks.test.ts > Think scheduled tasks > does not arm a keepAlive heartbeat on alarm() when no workflow notifications are pending (#1703)",
    "streaming-message-id.test.ts > Think — streaming assistant message-id alignment > stamps the persisted assistant id onto a new turn's start chunk",
    "submissions.test.ts > Think durable submissions > completes multiple submissions in FIFO order",
    "submissions.test.ts > Think durable submissions > deduplicates by submission id",
    "submissions.test.ts > Think durable submissions > deduplicates concurrent first submissions with the same idempotency key",
    "submissions.test.ts > Think durable submissions > deduplicates retries by idempotency key without appending duplicate messages",
    "submissions.test.ts > Think durable submissions > does not treat client body workflow-shaped data as workflow configuration",
    "submissions.test.ts > Think durable submissions > does not treat public workflow-shaped metadata as workflow configuration",
    "submissions.test.ts > Think durable submissions > stores error status and message when an in-band stream error occurs",
    "submissions.test.ts > Think durable submissions > stores error status and message when turn setup throws",
    "telegram-messenger.test.ts > telegram messenger adapter config (#2397) > allows unverified adapter webhooks only when Think verifies them",
    "telegram-messenger.test.ts > telegram messenger adapter config (#2397) > forwards nativeStreaming to the adapter",
    "think-eviction.test.ts > Think recovery after forced Durable Object eviction > rebuilds the in-memory transcript cache without running a turn",
    "think-session.test.ts > Think — Session integration > does not append missing messages to cache on direct session updateMessage calls",
    "think-session.test.ts > Think — Session integration > keeps cache aligned for direct session appendMessage calls",
    "think-session.test.ts > Think — Session integration > provides a cache-aware append helper for subclasses",
    "think-session.test.ts > Think — Session integration > returns a copy from getMessages",
    "think-session.test.ts > Think — Session integration > should use tree-structured messages via Session",
    "think-session.test.ts > Think — abort > should expose chat request ids for cross-RPC cancellation",
    "think-session.test.ts > Think — abort > should stop streaming on abort and not call onDone",
    "think-session.test.ts > Think — addMessages > appends without starting a model turn",
    "think-session.test.ts > Think — addMessages > chains an array linearly into one path",
    "think-session.test.ts > Think — addMessages > does not re-parent on upsert",
    "think-session.test.ts > Think — addMessages > is a no-op for empty input",
    "think-session.test.ts > Think — addMessages > is idempotent by message id in append mode",
    "think-session.test.ts > Think — addMessages > keeps the live cache coherent",
    "think-session.test.ts > Think — addMessages > keeps the live view correct when an explicit parentId branches",
    "think-session.test.ts > Think — addMessages > persists durably and coherently when called mid-turn",
    "think-session.test.ts > Think — addMessages > updates in place in upsert mode",
    "think-session.test.ts > Think — async configureSession > should initialize and chat with async configureSession",
    "think-session.test.ts > Think — async configureSession > should support multiple turns after async init",
    "think-session.test.ts > Think — body persistence > should default to undefined when no body set",
    "think-session.test.ts > Think — chatRecovery > chat turn with recovery=true works normally and cleans up fibers",
    "think-session.test.ts > Think — chatRecovery > multiple sequential turns don't leak fibers",
    "think-session.test.ts > Think — chatRecovery > saveMessages with recovery wraps in fiber and cleans up",
    "think-session.test.ts > Think — context blocks > does not warn when classifyChatError is overridden",
    "think-session.test.ts > Think — context blocks > should configure session with context blocks",
    "think-session.test.ts > Think — continueLastTurn > should set continuation: true on continueLastTurn",
    "think-session.test.ts > Think — core > should accumulate messages across multiple turns",
    "think-session.test.ts > Think — core > should broadcast RPC chat message updates to connected clients",
    "think-session.test.ts > Think — core > should broadcast programmatic clearMessages to connected clients",
    "think-session.test.ts > Think — core > should build assistant message with text parts",
    "think-session.test.ts > Think — core > should clear all messages",
    "think-session.test.ts > Think — core > should forward routed props to onStart",
    "think-session.test.ts > Think — core > should return empty messages before first chat",
    "think-session.test.ts > Think — error handling > bounds a turn that streams a chunk and stalls on every attempt",
    "think-session.test.ts > Think — error handling > does not retry an aborted turn under a transient classifier (inStream: false)",
    "think-session.test.ts > Think — error handling > does not retry an aborted turn under a transient classifier (inStream: true)",
    "think-session.test.ts > Think — error handling > never schedules live recovery for a Durable Object 'code-update-reset' (inStream: false)",
    "think-session.test.ts > Think — error handling > never schedules live recovery for a Durable Object 'code-update-reset' (inStream: true)",
    "think-session.test.ts > Think — error handling > never schedules live recovery for a Durable Object 'storage-reset' (inStream: false)",
    "think-session.test.ts > Think — error handling > should persist partial assistant message on error",
    "think-session.test.ts > Think — error handling > surfaces the stall as a terminal error when onChatRecovery throws",
    "think-session.test.ts > Think — getConfig inside configureSession > should not throw when getConfig() is called in configureSession on first start",
    "think-session.test.ts > Think — onChatRecovery > does not duplicate an already-persisted completed stream on recovery",
    "think-session.test.ts > Think — onChatRecovery > { continue: false } persists but does not schedule continuation",
    "think-session.test.ts > Think — onChatRecovery > { persist: false, continue: false } skips both",
    "think-session.test.ts > Think — pre-Sessions session API keeps working > accepts the positional appendMessage / getHistory / getRecentHistory forms",
    "think-session.test.ts > Think — pre-Sessions session API keeps working > routes a failing compaction function through onCompactionError()",
    "think-session.test.ts > Think — row size enforcement > should compact large tool outputs",
    "think-session.test.ts > Think — row size enforcement > should pass through small messages unchanged",
    "think-session.test.ts > Think — row size enforcement > should truncate large text parts for non-assistant messages",
    "think-session.test.ts > Think — sanitization > should filter empty reasoning parts without providerMetadata",
    "think-session.test.ts > Think — sanitization > should pass through messages without OpenAI metadata unchanged",
    "think-session.test.ts > Think — sanitization > should preserve reasoning parts with providerMetadata",
    "think-session.test.ts > Think — sanitization > should strip OpenAI ephemeral itemId from providerMetadata",
    "think-session.test.ts > Think — sanitization > should strip reasoningEncryptedContent from OpenAI metadata",
    "think-session.test.ts > Think — saveMessages > should broadcast to connected clients",
    "think-session.test.ts > Think — saveMessages > should fire onChatResponse",
    "think-session.test.ts > Think — saveMessages > should inject messages and run a turn",
    "think-session.test.ts > Think — saveMessages > should support function form",
    "think-session.test.ts > Think — saveMessages with external AbortSignal > runs to completion when the signal is never aborted",
    "think-session.test.ts > Think — tool call preservation > preserves earlier client-side tool results across turns",
    "think-session.test.ts > Think — waitUntilStable > detects a pending CLIENT-tool interaction (the SPA can still replay the tool-result)",
    "think-session.test.ts > Think — waitUntilStable > detects pending approval",
    "think-session.test.ts > Think — waitUntilStable > does NOT report pending for a dead SERVER-tool input-available in the accumulator",
    "think-session.test.ts > Think — waitUntilStable > returns false when no pending after tool result applied",
    "think-session.test.ts > Think — waitUntilStable > stops reporting pending once the accumulator's client-tool part resolves",
    "turn-metadata.test.ts > turn metadata > returns undefined when the latest user message carries none",
    "workflows.test.ts > ThinkWorkflow > prompt step RPC disposal > disposes submitMessages and waitForEvent results after copying serializable data",
    "workflows.test.ts > ThinkWorkflow > prompt step RPC disposal > disposes the submitMessages result before cancelling a timed-out prompt",
    "workflows.test.ts > ThinkWorkflow > prompt step RPC disposal > keeps the waitForEvent result alive while validating nested output",
    "ws-web-channel.test.ts > Think — WebSocket chat runs on the web channel (#2255) > still runs server-driven turns without a channel by default"
  ]
}