@yanlinglabs/winter-conformance 0.0.16 → 0.0.20

package/dist/official/web-tools-script.d.ts

@@ -0,0 +1,118 @@
+import { type OfficialRoots, type RawFrame, type SseEvent } from "./differential-harness.js";
+/** One scripted search hit. `extra` rides along on the WIRE only (page_age, encrypted_content, ...) to prove it never survives. */
+export interface ScriptedHit {
+    title: string;
+    url: string;
+    extra?: Record<string, unknown>;
+}
+/**
+ * The claude-side block sequence an inner `web_search_20250305` response carries. A `search` is a
+ * `server_tool_use` block immediately followed by its `web_search_tool_result` block (hits, or a
+ * result-block error).
+ */
+export type ScriptedSearchBlock = {
+    kind: "text";
+    text: string;
+} | {
+    kind: "search";
+    query: string;
+    hits: ScriptedHit[];
+} | {
+    kind: "search_error";
+    query: string;
+    errorCode: string;
+};
+/** Structurally identical to the runtime's own `WebSearchStreamEvent` (declared here rather than imported so this module stays free of cross-package imports; each test file passes the result straight to the real assembler, so a drift in that type is a compile error THERE). */
+export type WinterSearchEvent = {
+    type: "text";
+    text: string;
+} | {
+    type: "search_result";
+    hits: Array<{
+        title: string;
+        url: string;
+    }>;
+} | {
+    type: "search_error";
+    code: string;
+};
+/**
+ * The SAME scripted sequence in Winter's event shape. Deliberately passes every `extra` field
+ * through on the hit objects (as untyped excess properties), so "only title and url survive" is
+ * proven on Winter's side by the assembler dropping them, not by this mapper never offering them.
+ */
+export declare function toWinterEvents(blocks: readonly ScriptedSearchBlock[]): WinterSearchEvent[];
+/** The inner response as an SSE stream: one content block per text, two per search. */
+export declare function sseInnerSearchTurn(blocks: readonly ScriptedSearchBlock[], model?: string): SseEvent[];
+export declare const INNER_SEARCH_USER_PREFIX = "Perform a web search for the query: ";
+export declare const INNER_FETCH_USER_PREFIX = "\nWeb page content:\n";
+export declare function toolsOf(body: RawFrame): RawFrame[];
+/** The inner search call: recognised by the server tool in its tool list, never by position. */
+export declare function isInnerSearchRequest(body: RawFrame): boolean;
+/** The first user message's text, whether the wire carries it as a string or as a one-block array. */
+export declare function firstUserText(body: RawFrame): string;
+/** The inner digest call: no tools at all, and the page-content template opening the user text. */
+export declare function isInnerFetchRequest(body: RawFrame): boolean;
+export declare function hasToolResult(messages: readonly RawFrame[]): boolean;
+export interface OfficialToolResult {
+    /** The tool_result `content` as the binary's own stdout `user` frame carries it -- the tool's PURE output. */
+    content: string;
+    isError: boolean;
+    /** The binary's structured result for the call (`tool_use_result` on the same frame), when it has one. */
+    structured: unknown;
+}
+/** Every tool_result the binary emitted on stdout, keyed by `tool_use_id` (never by position -- batched calls finish in any order). */
+export declare function toolResultsFromFrames(frames: readonly RawFrame[]): Map<string, OfficialToolResult>;
+/** Every tool_result in a captured main-loop REQUEST, keyed by `tool_use_id` -- what actually went back to the API. */
+export declare function toolResultsFromRequest(body: RawFrame): Map<string, string>;
+/**
+ * The binary's MAIN LOOP appends a token-budget note to a tool_result on the wire (observed on the
+ * last tool_result of a request). It is loop decoration, not part of any tool's own output -- the
+ * stdout frame for the same call does not carry it -- so the wire comparison allows exactly this
+ * suffix and nothing else.
+ */
+export declare const MAIN_LOOP_WIRE_SUFFIX: RegExp;
+/** True when `wire` is `pure` exactly, or `pure` plus the one known main-loop suffix. */
+export declare function wireMatchesPure(wire: string, pure: string): boolean;
+export interface ProxyTrap {
+    url: string;
+    /** The first line of every request that reached the trap -- must stay EMPTY for a hermetic run. */
+    hits: string[];
+    stop(): void;
+}
+export declare function startProxyTrap(): ProxyTrap;
+export interface LoopbackTls {
+    key: string;
+    cert: string;
+    certPath: string;
+}
+/** A self-signed cert for 127.0.0.1, written under `dir` (a mkdtemp root the caller owns). Returns `{reason}` when openssl is unavailable, so the caller can skip plainly rather than fail obscurely. */
+export declare function makeLoopbackTls(dir: string): LoopbackTls | {
+    reason: string;
+};
+export interface OfficialRun {
+    /** Every POST body the loopback received, in arrival order. */
+    requests: RawFrame[];
+    /** Every stdout frame the binary emitted. */
+    frames: RawFrame[];
+    /** First lines of anything that tried to leave the box (see the header) -- asserted empty by every caller. */
+    trapHits: string[];
+}
+export interface OfficialRunOptions {
+    binaryPath: string;
+    model?: string;
+    prompt: string;
+    route: (messages: RawFrame[], body: RawFrame, count: number) => Response;
+    logPrefix: string;
+    /** Extra `--settings` JSON (always merged over `skipWebFetchPreflight: true`). */
+    settings?: Record<string, unknown>;
+    extraEnv?: Record<string, string>;
+    /** Called with the roots before the spawn -- for a caller that needs files under them (a TLS cert). */
+    roots?: OfficialRoots;
+}
+/**
+ * One headless `-p` invocation of the pinned binary against a scripted loopback, permission mode
+ * `bypassPermissions` -- the SAME way every other file in this family lets a tool run (WebSearch and
+ * WebFetch both prompt otherwise); no allow rule, no hook, nothing more invasive than that.
+ */
+export declare function runOfficialOnce(opts: OfficialRunOptions): Promise<OfficialRun>;

package/goldens/advertised-set-round.trace.json

@@ -38,6 +38,8 @@
         "TaskOutput",
         "TaskStop",
         "TaskUpdate",
+        "WebFetch",
+        "WebSearch",
         "Workflow",
         "Write"
       ],
@@ -81,7 +83,39 @@
       "subtype": "success",
       "is_error": false,
       "result": "echo: <system-reminder>\nAvailable agent types for the Agent tool:\n- claude: Catch-all for any task that doesn't fit a more specific agent. (Tools: *)\n- Explore: Fast read-only search agent for locating code. Use it to find files by pattern (eg. \"src/components/**/*.tsx\"), grep for symbols or keywords (eg. \"API endpoints\"), or answer \"where is X defined / which files reference Y.\" Do NOT use it for code review, design-doc auditing, cross-file consistency checks, or open-ended analysis — it reads excerpts rather than whole files and will miss content past its read window. When calling, specify search breadth: \"quick\" for a single targeted lookup, \"medium\" for moderate exploration, or \"very thorough\" to search across multiple locations and naming conventions. (Tools: All tools except Agent, Artifact, ExitPlanMode, Edit, Write, NotebookEdit)\n- general-purpose: General-purpose agent for researching complex questions, searching for code, and executing multi-step tasks. When you are searching for a keyword or file and are not confident that you will find the right match in the first few tries use this agent to perform the search for you. (Tools: *)\n- Plan: Software architect agent for designing implementation plans. Use this when you need to plan the implementation strategy for a task. Returns step-by-step plans, identifies critical files, and considers architectural trade-offs. (Tools: All tools except Agent, Artifact, ExitPlanMode, Edit, Write, NotebookEdit)\n\nWhen you launch multiple agents for independent work, send them in a single message with multiple tool uses so they run concurrently.\n</system-reminder>\n\n<system-reminder>\nAs you answer the user's questions, you can use the following context:\n# currentDate\nToday's date is <FIXTURE-DATE>.\n\n      IMPORTANT: this context may or may not be relevant to your tasks. You should not respond to this context unless it is highly relevant to your task.\n</system-reminder>\n\n\nhi",
-      "permission_denials": []
+      "usage": {
+        "output_tokens_details": {
+          "thinking_tokens": 0
+        },
+        "input_tokens": "<input_tokens>",
+        "cache_creation_input_tokens": "<cache_creation_input_tokens>",
+        "cache_read_input_tokens": "<cache_read_input_tokens>",
+        "output_tokens": "<output_tokens>",
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 0,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "",
+        "iterations": [],
+        "speed": "standard"
+      },
+      "permission_denials": [],
+      "modelUsage": {
+        "winter-test/echo": {
+          "inputTokens": "<inputTokens>",
+          "outputTokens": "<outputTokens>",
+          "cacheReadInputTokens": "<cacheReadInputTokens>",
+          "cacheCreationInputTokens": "<cacheCreationInputTokens>",
+          "webSearchRequests": 0,
+          "costUSD": 0,
+          "canonicalModel": "winter-test/echo"
+        }
+      }
     }
   }
 ]

package/goldens/background-task-round.trace.json

@@ -39,6 +39,8 @@
         "TaskOutput",
         "TaskStop",
         "TaskUpdate",
+        "WebFetch",
+        "WebSearch",
         "Workflow",
         "Write"
       ],
@@ -210,7 +212,39 @@
       "subtype": "success",
       "is_error": false,
       "result": "backgrounded",
-      "permission_denials": []
+      "usage": {
+        "output_tokens_details": {
+          "thinking_tokens": 0
+        },
+        "input_tokens": "<input_tokens>",
+        "cache_creation_input_tokens": "<cache_creation_input_tokens>",
+        "cache_read_input_tokens": "<cache_read_input_tokens>",
+        "output_tokens": "<output_tokens>",
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 0,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "",
+        "iterations": [],
+        "speed": "standard"
+      },
+      "permission_denials": [],
+      "modelUsage": {
+        "winter-test/echo": {
+          "inputTokens": "<inputTokens>",
+          "outputTokens": "<outputTokens>",
+          "cacheReadInputTokens": "<cacheReadInputTokens>",
+          "cacheCreationInputTokens": "<cacheCreationInputTokens>",
+          "webSearchRequests": 0,
+          "costUSD": 0,
+          "canonicalModel": "winter-test/echo"
+        }
+      }
     }
   }
 ]

package/goldens/bash-background-round.trace.json

@@ -39,6 +39,8 @@
         "TaskOutput",
         "TaskStop",
         "TaskUpdate",
+        "WebFetch",
+        "WebSearch",
         "Workflow",
         "Write"
       ],
@@ -150,7 +152,39 @@
       "subtype": "success",
       "is_error": false,
       "result": "backgrounded",
-      "permission_denials": []
+      "usage": {
+        "output_tokens_details": {
+          "thinking_tokens": 0
+        },
+        "input_tokens": "<input_tokens>",
+        "cache_creation_input_tokens": "<cache_creation_input_tokens>",
+        "cache_read_input_tokens": "<cache_read_input_tokens>",
+        "output_tokens": "<output_tokens>",
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 0,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "",
+        "iterations": [],
+        "speed": "standard"
+      },
+      "permission_denials": [],
+      "modelUsage": {
+        "winter-test/echo": {
+          "inputTokens": "<inputTokens>",
+          "outputTokens": "<outputTokens>",
+          "cacheReadInputTokens": "<cacheReadInputTokens>",
+          "cacheCreationInputTokens": "<cacheCreationInputTokens>",
+          "webSearchRequests": 0,
+          "costUSD": 0,
+          "canonicalModel": "winter-test/echo"
+        }
+      }
     }
   }
 ]

package/goldens/canusetool-approved-round.trace.json

@@ -39,6 +39,8 @@
         "TaskOutput",
         "TaskStop",
         "TaskUpdate",
+        "WebFetch",
+        "WebSearch",
         "Workflow",
         "Write"
       ],
@@ -119,7 +121,39 @@
       "subtype": "success",
       "is_error": false,
       "result": "tool round done",
-      "permission_denials": []
+      "usage": {
+        "output_tokens_details": {
+          "thinking_tokens": 0
+        },
+        "input_tokens": "<input_tokens>",
+        "cache_creation_input_tokens": "<cache_creation_input_tokens>",
+        "cache_read_input_tokens": "<cache_read_input_tokens>",
+        "output_tokens": "<output_tokens>",
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 0,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "",
+        "iterations": [],
+        "speed": "standard"
+      },
+      "permission_denials": [],
+      "modelUsage": {
+        "winter-test/echo": {
+          "inputTokens": "<inputTokens>",
+          "outputTokens": "<outputTokens>",
+          "cacheReadInputTokens": "<cacheReadInputTokens>",
+          "cacheCreationInputTokens": "<cacheCreationInputTokens>",
+          "webSearchRequests": 0,
+          "costUSD": 0,
+          "canonicalModel": "winter-test/echo"
+        }
+      }
     }
   }
 ]

package/goldens/compaction-auto-round.trace.json

@@ -39,6 +39,8 @@
         "TaskOutput",
         "TaskStop",
         "TaskUpdate",
+        "WebFetch",
+        "WebSearch",
         "Workflow",
         "Write"
       ],
@@ -82,7 +84,39 @@
       "subtype": "success",
       "is_error": false,
       "result": "reply 1",
-      "permission_denials": []
+      "usage": {
+        "output_tokens_details": {
+          "thinking_tokens": 0
+        },
+        "input_tokens": "<input_tokens>",
+        "cache_creation_input_tokens": "<cache_creation_input_tokens>",
+        "cache_read_input_tokens": "<cache_read_input_tokens>",
+        "output_tokens": "<output_tokens>",
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 0,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "",
+        "iterations": [],
+        "speed": "standard"
+      },
+      "permission_denials": [],
+      "modelUsage": {
+        "winter-test/echo": {
+          "inputTokens": "<inputTokens>",
+          "outputTokens": "<outputTokens>",
+          "cacheReadInputTokens": "<cacheReadInputTokens>",
+          "cacheCreationInputTokens": "<cacheCreationInputTokens>",
+          "webSearchRequests": 0,
+          "costUSD": 0,
+          "canonicalModel": "winter-test/echo"
+        }
+      }
     }
   },
   {
@@ -110,7 +144,39 @@
       "subtype": "success",
       "is_error": false,
       "result": "reply 2",
-      "permission_denials": []
+      "usage": {
+        "output_tokens_details": {
+          "thinking_tokens": 0
+        },
+        "input_tokens": "<input_tokens>",
+        "cache_creation_input_tokens": "<cache_creation_input_tokens>",
+        "cache_read_input_tokens": "<cache_read_input_tokens>",
+        "output_tokens": "<output_tokens>",
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 0,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "",
+        "iterations": [],
+        "speed": "standard"
+      },
+      "permission_denials": [],
+      "modelUsage": {
+        "winter-test/echo": {
+          "inputTokens": "<inputTokens>",
+          "outputTokens": "<outputTokens>",
+          "cacheReadInputTokens": "<cacheReadInputTokens>",
+          "cacheCreationInputTokens": "<cacheCreationInputTokens>",
+          "webSearchRequests": 0,
+          "costUSD": 0,
+          "canonicalModel": "winter-test/echo"
+        }
+      }
     }
   },
   {
@@ -138,7 +204,39 @@
       "subtype": "success",
       "is_error": false,
       "result": "reply 3",
-      "permission_denials": []
+      "usage": {
+        "output_tokens_details": {
+          "thinking_tokens": 0
+        },
+        "input_tokens": "<input_tokens>",
+        "cache_creation_input_tokens": "<cache_creation_input_tokens>",
+        "cache_read_input_tokens": "<cache_read_input_tokens>",
+        "output_tokens": "<output_tokens>",
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 0,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "",
+        "iterations": [],
+        "speed": "standard"
+      },
+      "permission_denials": [],
+      "modelUsage": {
+        "winter-test/echo": {
+          "inputTokens": "<inputTokens>",
+          "outputTokens": "<outputTokens>",
+          "cacheReadInputTokens": "<cacheReadInputTokens>",
+          "cacheCreationInputTokens": "<cacheCreationInputTokens>",
+          "webSearchRequests": 0,
+          "costUSD": 0,
+          "canonicalModel": "winter-test/echo"
+        }
+      }
     }
   },
   {
@@ -166,7 +264,39 @@
       "subtype": "success",
       "is_error": false,
       "result": "reply 4",
-      "permission_denials": []
+      "usage": {
+        "output_tokens_details": {
+          "thinking_tokens": 0
+        },
+        "input_tokens": "<input_tokens>",
+        "cache_creation_input_tokens": "<cache_creation_input_tokens>",
+        "cache_read_input_tokens": "<cache_read_input_tokens>",
+        "output_tokens": "<output_tokens>",
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 0,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "",
+        "iterations": [],
+        "speed": "standard"
+      },
+      "permission_denials": [],
+      "modelUsage": {
+        "winter-test/echo": {
+          "inputTokens": "<inputTokens>",
+          "outputTokens": "<outputTokens>",
+          "cacheReadInputTokens": "<cacheReadInputTokens>",
+          "cacheCreationInputTokens": "<cacheCreationInputTokens>",
+          "webSearchRequests": 0,
+          "costUSD": 0,
+          "canonicalModel": "winter-test/echo"
+        }
+      }
     }
   },
   {
@@ -219,7 +349,39 @@
       "subtype": "success",
       "is_error": false,
       "result": "reply 5",
-      "permission_denials": []
+      "usage": {
+        "output_tokens_details": {
+          "thinking_tokens": 0
+        },
+        "input_tokens": "<input_tokens>",
+        "cache_creation_input_tokens": "<cache_creation_input_tokens>",
+        "cache_read_input_tokens": "<cache_read_input_tokens>",
+        "output_tokens": "<output_tokens>",
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 0,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "",
+        "iterations": [],
+        "speed": "standard"
+      },
+      "permission_denials": [],
+      "modelUsage": {
+        "winter-test/echo": {
+          "inputTokens": "<inputTokens>",
+          "outputTokens": "<outputTokens>",
+          "cacheReadInputTokens": "<cacheReadInputTokens>",
+          "cacheCreationInputTokens": "<cacheCreationInputTokens>",
+          "webSearchRequests": 0,
+          "costUSD": 0,
+          "canonicalModel": "winter-test/echo"
+        }
+      }
     }
   }
 ]