@selesai/code 0.13.2 → 0.13.4

package/CHANGELOG.md

@@ -2,6 +2,26 @@
 
 All notable changes to `@selesai/code` will be documented in this file.
 
+## [0.13.4] - 2026-08-31
+
+### Added
+- **Bundled `pi-subagents` 0.60.0.** `subagent({ action: "list", capabilities: true })` returns a compact prompt-free capability catalog (`details.catalog`: agent name, description, source, runner, tools/model/execution/output/extensions snapshot, executable and restriction status), and `list` results stay human text. Forked children now stamp an OpenAI-compatible `prompt_cache_key` (`pi-fork:<sha256>`) on provider requests for prompt-cache reuse; timeout recovery evidence (report status, changed tracked files, dirty-worktree classification) is projected into async status and `subagent_wait` completions; `runs.lanes` lane plans (stages, phase, label, structured output) are emitted and rendered as workflow-graph stages; `runs.host` rejects per-step `cwd` with a hint; watchdog repo signatures skip untracked scans for home-root and entry-less repos; single-output snapshots propagate non-ENOENT stat errors; background processes use `windowsHide` and detached-only-on-unix.
+- **Bundled `pi-web-agent` 1.10.0.** Search backends: You.com (`YDC_API_KEY`), Exa (`EXA_API_KEY`), Tavily (`TAVILY_API_KEY`) in addition to DuckDuckGo, SearXNG and Brave; search fanout across configured providers with dedupe/agreement ranking (`backends.search.fanout` with `off`/`on`/`auto` modes and provider selection); keyless Tavily fallback when DuckDuckGo is bot-blocked (disable with `PI_WEB_AGENT_DISABLE_KEYLESS_FALLBACK=1`); DuckDuckGo hardened against bot-walls (browser headers + one retry); GitHub/PDF/YouTube direct readers wired into the fetch path; canonical URL normalization (tracking-param stripping); the model tool result now carries the full synthesized findings/sources/caveats regardless of terminal presentation.
+- **GLM-5.3 models in the default catalog.** Added `glm-5.3` (GLM-5.3) to the bundled `tokenin` provider and upgraded `glm-5.3-flash` to text+image input with a full `thinkingLevelMap`.
+
+### Changed
+- **Cost reconciliation keys every streamed response id.** Streamed bodies repeat the response id across chunks and carry tool-call ids (`call_*`), so the fetch wrapper now keys the billed cost by every id found and `message_end` consumes only the one matching the finalized message's `responseId`; a duplicate capture (retry) is marked ambiguous instead of assigning either bill.
+- **Bundled `pi-intercom` 0.12.1.** During a turn triggered by an inbound ask, a non-`reply` `send` to a different target is refused so guessed parent/root CWD cannot receive an accidental reply; the `toolVisibility` setting and `after-first-use` reveal path were removed (the generic `intercom` tool stays in the active tool set for prompt-cache stability; obsolete `toolVisibility` config keys are ignored).
+
+## [0.13.3] - 2026-08-31
+
+### Added
+- **Token-In multi-key auto-failover.** The `tokenin-onboarding` extension now registers a `tokenin` provider whose `streamSimple` automatically rotates to an alternative saved account when the active key fails with a 401, 429, or budget/quota error. Failed keys go on a 5-minute in-memory cooldown before being retried, and the newly active credential is persisted to both `tokenin-auth.json` and `auth.json`.
+
+### Changed
+- **Cost reconciliation replaces the finalized assistant cost.** The `cost-reconcile` fetch wrapper now finishes capturing the provider-reported cost before the provider SDK finalizes its assistant message, so `message_end` replaces `usage.cost.total` with the billed amount before the message is persisted. Providers whose payloads carry no cost keep their rate-card estimate.
+- **Default model catalog reasoning maps.** Added `thinkingLevelMap` entries (and `low`=low / `max` mappings) across bundled model defaults and enabled reasoning on Qwen3.8 27B and Gemini 3.7 Flash so thinking budgets map correctly.
+
 ## [0.13.2] - 2026-08-30
 
 ### Added

package/dist/defaults/models.json

@@ -19,7 +19,7 @@
           "maxTokens": 64000,
           "thinkingLevelMap": {
             "minimal": null,
-            "low": null,
+            "low": "low",
             "medium": null,
             "high": "high",
             "xhigh": null,
@@ -38,7 +38,7 @@
           "maxTokens": 64000,
           "thinkingLevelMap": {
             "minimal": null,
-            "low": null,
+            "low": "low",
             "medium": null,
             "high": "high",
             "xhigh": null,
@@ -57,7 +57,7 @@
           "maxTokens": 64000,
           "thinkingLevelMap": {
             "minimal": null,
-            "low": null,
+            "low": "low",
             "medium": null,
             "high": "high",
             "xhigh": null,
@@ -114,7 +114,7 @@
           "maxTokens": 64000,
           "thinkingLevelMap": {
             "minimal": null,
-            "low": null,
+            "low": "low",
             "medium": null,
             "high": "high",
             "xhigh": null,
@@ -130,7 +130,15 @@
           "reasoning": true,
           "input": ["text", "image"],
           "contextWindow": 512000,
-          "maxTokens": 128000
+          "maxTokens": 128000,
+          "thinkingLevelMap": {
+            "minimal": null,
+            "low": "low",
+            "medium": null,
+            "high": "high",
+            "xhigh": null,
+            "max": "max"
+          }
         },
         {
           "id": "gpt-5.6-terra",
@@ -138,7 +146,15 @@
           "reasoning": true,
           "input": ["text", "image"],
           "contextWindow": 512000,
-          "maxTokens": 128000
+          "maxTokens": 128000,
+          "thinkingLevelMap": {
+            "minimal": null,
+            "low": "low",
+            "medium": null,
+            "high": "high",
+            "xhigh": null,
+            "max": null
+          }
         },
         {
           "id": "gpt-5.6-sol",
@@ -146,7 +162,15 @@
           "reasoning": true,
           "input": ["text", "image"],
           "contextWindow": 512000,
-          "maxTokens": 128000
+          "maxTokens": 128000,
+          "thinkingLevelMap": {
+            "minimal": null,
+            "low": "low",
+            "medium": null,
+            "high": "high",
+            "xhigh": null,
+            "max": null
+          }
         },
         {
           "id": "kimi-k3",
@@ -200,7 +224,7 @@
           "id": "glm-5.3-flash",
           "name": "GLM-5.3 Flash",
           "reasoning": true,
-          "input": ["text"],
+          "input": ["text", "image"],
           "contextWindow": 1000000,
           "maxTokens": 64000,
           "thinkingLevelMap": {
@@ -209,16 +233,40 @@
             "medium": null,
             "high": "high",
             "xhigh": null,
-            "max": null
+            "max": "max"
+          }
+        },
+        {
+          "id": "glm-5.3",
+          "name": "GLM-5.3",
+          "reasoning": true,
+          "input": ["text", "image"],
+          "contextWindow": 1000000,
+          "maxTokens": 64000,
+          "thinkingLevelMap": {
+            "minimal": null,
+            "low": "low",
+            "medium": null,
+            "high": "high",
+            "xhigh": null,
+            "max": "max"
           }
         },
         {
           "id": "qwen3.8-27b",
           "name": "Qwen3.8 27B",
-          "reasoning": false,
+          "reasoning": true,
           "input": ["text", "image"],
           "contextWindow": 256000,
           "maxTokens": 64000,
+          "thinkingLevelMap": {
+            "minimal": null,
+            "low": "low",
+            "medium": null,
+            "high": "high",
+            "xhigh": null,
+            "max": "max"
+          },
           "compat": {
             "thinkingFormat": "qwen"
           }
@@ -226,10 +274,18 @@
         {
           "id": "gemini-3.7-flash",
           "name": "Gemini 3.7 Flash",
-          "reasoning": false,
-          "input": ["text", "image" ],
+          "reasoning": true,
+          "input": ["text", "image"],
           "contextWindow": 512000,
-          "maxTokens": 64000
+          "maxTokens": 64000,
+          "thinkingLevelMap": {
+            "minimal": null,
+            "low": "low",
+            "medium": null,
+            "high": "high",
+            "xhigh": null,
+            "max": null
+          }
         },
         {
           "id": "Qwen3-VL",
@@ -238,6 +294,14 @@
           "input": ["text", "image"],
           "contextWindow": 256000,
           "maxTokens": 8192,
+          "thinkingLevelMap": {
+            "minimal": null,
+            "low": "low",
+            "medium": null,
+            "high": "high",
+            "xhigh": null,
+            "max": null
+          },
           "compat": {
             "supportsDeveloperRole": false,
             "supportsReasoningEffort": false,
@@ -251,6 +315,14 @@
           "input": ["text", "image"],
           "contextWindow": 256000,
           "maxTokens": 8192,
+          "thinkingLevelMap": {
+            "minimal": null,
+            "low": "low",
+            "medium": null,
+            "high": "high",
+            "xhigh": null,
+            "max": null
+          },
           "compat": {
             "supportsDeveloperRole": false,
             "supportsReasoningEffort": false,

package/dist/extensions/cost-reconcile.test.ts

@@ -1,5 +1,63 @@
-import { describe, expect, it } from "vitest";
-import { extractCosts, parseCostHeader } from "./cost-reconcile.ts";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+import costReconcileExtension, { extractCosts, parseCostHeader } from "./cost-reconcile.ts";
+
+type Handler = (event: any, ctx: any) => any;
+
+const FETCH_PATCHED = Symbol.for("selesai.cost-reconcile.fetch-patched");
+let originalFetch: typeof globalThis.fetch;
+
+function createHarness() {
+	const handlers = new Map<string, Handler>();
+	const appendEntry = vi.fn();
+	const pi = {
+		on: vi.fn((event: string, handler: Handler) => handlers.set(event, handler)),
+		appendEntry,
+	};
+	costReconcileExtension(pi as any);
+	return { appendEntry, handlers };
+}
+
+async function startSession(handlers: Map<string, Handler>): Promise<void> {
+	await handlers.get("session_start")!({}, { sessionManager: { getEntries: () => [] } });
+}
+
+function assistantMessage(responseId: string, total = 0.75) {
+	return {
+		role: "assistant",
+		provider: "openrouter",
+		model: "openai/gpt-4.1-mini",
+		responseModel: "openai/gpt-4.1-mini",
+		responseId,
+		stopReason: "stop",
+		content: [],
+		usage: {
+			input: 10,
+			output: 5,
+			cacheRead: 2,
+			cacheWrite: 1,
+			cost: { input: 0.2, output: 0.4, cacheRead: 0.1, cacheWrite: 0.05, total },
+		},
+	};
+}
+
+function installFetch(body: string, headers: Record<string, string> = {}) {
+	const upstreamFetch = vi.fn(async () =>
+		new Response(body, { headers: { "content-type": "text/event-stream", ...headers } }),
+	);
+	globalThis.fetch = upstreamFetch as typeof globalThis.fetch;
+	return upstreamFetch;
+}
+
+beforeEach(() => {
+	originalFetch = globalThis.fetch;
+	delete (globalThis as typeof globalThis & { [FETCH_PATCHED]?: boolean })[FETCH_PATCHED];
+});
+
+afterEach(() => {
+	globalThis.fetch = originalFetch;
+	delete (globalThis as typeof globalThis & { [FETCH_PATCHED]?: boolean })[FETCH_PATCHED];
+	vi.restoreAllMocks();
+});
 
 describe("parseCostHeader", () => {
 	it("parses LiteLLM scientific-notation cost", () => {
@@ -10,11 +68,17 @@ describe("parseCostHeader", () => {
 		expect(parseCostHeader("0.00123")).toBe(0.00123);
 	});
 
-	it("rejects null, garbage, and negatives", () => {
+	it("rejects null, blank, garbage, and negative values", () => {
 		expect(parseCostHeader(null)).toBeUndefined();
+		expect(parseCostHeader("")).toBeUndefined();
+		expect(parseCostHeader(" \t ")).toBeUndefined();
 		expect(parseCostHeader("abc")).toBeUndefined();
 		expect(parseCostHeader("-1")).toBeUndefined();
 	});
+
+	it("trims valid zero-valued headers", () => {
+		expect(parseCostHeader(" 0 ")).toBe(0);
+	});
 });
 
 describe("extractCosts", () => {
@@ -73,4 +137,136 @@ describe("extractCosts", () => {
 		const { ids } = extractCosts(body);
 		expect(ids.length).toBe(50);
 	});
-});
+});
+
+describe("live assistant usage reconciliation", () => {
+	it("replaces finalized assistant usage only after the streamed body completes and keeps the custom entry", async () => {
+		const responseId = "live-reconcile-1";
+		const body = `data: {"id":"${responseId}","usage":{"cost":0.0123}}\n\ndata: [DONE]\n\n`;
+		installFetch(body);
+		const { appendEntry, handlers } = createHarness();
+		await startSession(handlers);
+
+		const response = await globalThis.fetch("https://openrouter.ai/api/v1/chat/completions");
+		const messageEnd = handlers.get("message_end")!;
+		expect(await messageEnd({ message: assistantMessage(responseId) }, {})).toBeUndefined();
+
+		expect(await response.text()).toBe(body);
+		const result = await messageEnd({ message: assistantMessage(responseId) }, {});
+		expect(result.message.usage.cost).toEqual({
+			input: 0.2,
+			output: 0.4,
+			cacheRead: 0.1,
+			cacheWrite: 0.05,
+			total: 0.0123,
+		});
+		expect(appendEntry).toHaveBeenCalledWith(
+			"cost-reconcile",
+			expect.objectContaining({ provider: "openrouter", responseId, cost: 0.0123, source: "payload" }),
+		);
+	});
+
+	it("uses a valid zero-valued billed header over the payload cost", async () => {
+		const responseId = "live-header-zero-2";
+		installFetch(`data: {"id":"${responseId}","usage":{"cost":0.0123}}\n\n`, {
+			"x-litellm-response-cost": "0",
+		});
+		const { handlers } = createHarness();
+		await startSession(handlers);
+
+		const response = await globalThis.fetch("https://gateway.example/v1/chat/completions");
+		await response.text();
+		const result = await handlers.get("message_end")!({ message: assistantMessage(responseId) }, {});
+		expect(result.message.usage.cost.total).toBe(0);
+	});
+
+	it("reconciles a tool-call stream: keys cost by every id, message_end consumes the matching one", async () => {
+		const responseId = "live-toolcall-3";
+		const body =
+			`data: {"id":"${responseId}","choices":[{"delta":{"tool_calls":[{"id":"call_abc123","function":{"name":"bash","arguments":""}}]}}]}\n\n` +
+			`data: {"id":"${responseId}","usage":{"cost":0.0123}}\n\n` +
+			"data: [DONE]\n\n";
+		installFetch(body);
+		const { appendEntry, handlers } = createHarness();
+		await startSession(handlers);
+
+		const response = await globalThis.fetch("https://gateway.example/v1/chat/completions");
+		await response.text();
+		const message = assistantMessage(responseId);
+		const result = await handlers.get("message_end")!({ message }, {});
+		expect(result.message.usage.cost.total).toBe(0.0123);
+		expect(appendEntry).toHaveBeenCalledWith(
+			"cost-reconcile",
+			expect.objectContaining({ provider: "openrouter", responseId, cost: 0.0123, source: "payload" }),
+		);
+	});
+
+	it("does not apply a captured cost to a message whose id was not in the body", async () => {
+		const body = `data: {"id":"other-response","usage":{"cost":0.0123}}\n\ndata: [DONE]\n\n`;
+		installFetch(body);
+		const { appendEntry, handlers } = createHarness();
+		await startSession(handlers);
+
+		const response = await globalThis.fetch("https://gateway.example/v1/chat/completions");
+		await response.text();
+		const message = assistantMessage("unrelated-message");
+		expect(await handlers.get("message_end")!({ message }, {})).toBeUndefined();
+		expect(message.usage.cost.total).toBe(0.75);
+		expect(appendEntry).not.toHaveBeenCalled();
+	});
+
+	it("falls back when the process cache sees a duplicate response id", async () => {
+		const responseId = "live-duplicate-response-4";
+		installFetch(`data: {"id":"${responseId}","usage":{"cost":0.0123}}\n\n`);
+		const { appendEntry, handlers } = createHarness();
+		await startSession(handlers);
+
+		const first = await globalThis.fetch("https://gateway.example/v1/chat/completions");
+		const second = await globalThis.fetch("https://gateway.example/v1/chat/completions");
+		await Promise.all([first.text(), second.text()]);
+		const message = assistantMessage(responseId);
+		expect(await handlers.get("message_end")!({ message }, {})).toBeUndefined();
+		expect(message.usage.cost.total).toBe(0.75);
+		expect(appendEntry).not.toHaveBeenCalled();
+	});
+
+	it("preserves the original message when no valid billed cost was captured", async () => {
+		const responseId = "live-no-cost-3";
+		installFetch(`data: {"id":"${responseId}","usage":{"cost":-1}}\n\n`);
+		const { appendEntry, handlers } = createHarness();
+		await startSession(handlers);
+
+		const response = await globalThis.fetch("https://gateway.example/v1/chat/completions");
+		await response.text();
+		const message = assistantMessage(responseId);
+		expect(await handlers.get("message_end")!({ message }, {})).toBeUndefined();
+		expect(message.usage.cost.total).toBe(0.75);
+		expect(appendEntry).not.toHaveBeenCalled();
+	});
+
+	it("preserves terminal error messages even when a billed cost was captured", async () => {
+		const responseId = "live-terminal-4";
+		installFetch(`data: {"id":"${responseId}","usage":{"cost":0.0123}}\n\n`);
+		const { appendEntry, handlers } = createHarness();
+		await startSession(handlers);
+
+		const response = await globalThis.fetch("https://gateway.example/v1/chat/completions");
+		await response.text();
+		const message = { ...assistantMessage(responseId), stopReason: "error" };
+		expect(await handlers.get("message_end")!({ message }, {})).toBeUndefined();
+		expect(message.usage.cost.total).toBe(0.75);
+		expect(appendEntry).not.toHaveBeenCalled();
+	});
+
+	it("does not capture unrelated fetches", async () => {
+		const responseId = "unrelated-fetch-5";
+		const body = `data: {"id":"${responseId}","usage":{"cost":0.0123}}\n\n`;
+		installFetch(body);
+		const { handlers } = createHarness();
+		await startSession(handlers);
+
+		const response = await globalThis.fetch("https://example.com/data.json");
+		expect(await response.text()).toBe(body);
+		expect(await handlers.get("message_end")!({ message: assistantMessage(responseId) }, {})).toBeUndefined();
+	});
+});