@oh-my-pi/pi-ai 18.2.9 → 18.2.11

package/CHANGELOG.md

@@ -2,6 +2,13 @@
 
 ## [Unreleased]
 
+## [18.2.11] - 2026-09-23
+
+### Fixed
+
+- Fixed Claude Opus 5.5 not applying a mid-session switch to high-effort reasoning when the session started without an explicit effort setting.
+- Fixed Alibaba Token Plan monthly quotas not appearing in usage reports or the status line.
+
 ## [18.2.9] - 2026-09-22
 
 ### Added
@@ -2274,17 +2281,4 @@
 - Preserved Anthropic `stop_details` on assistant messages so refusal and sensitive classifier stops remain structurally visible to callers. ([#2290](https://github.com/can1357/oh-my-pi/issues/2290))
 - Fixed OpenAI Responses, Azure OpenAI Responses, and OpenAI Completions streams hanging until the 120s idle watchdog errored the turn when a provider delivers the terminal frame but never sends `[DONE]` nor closes the connection. `processResponsesStream` now breaks out of the event loop on `response.completed`/`response.incomplete` (mirroring the Codex websocket/SSE terminal break), and the completions consumer breaks once `finish_reason` plus a usage payload arrived — or, for hosts that never send usage, ends the stream cleanly via a short post-finish grace window (`iterateWithTerminalGrace`) that aborts the transport to release the socket.
 
-## [15.11.0] - 2026-06-10
-
-### Added
-
-- Added optional `ImageContent.detail` (`"auto" | "low" | "high" | "original"`): an OpenAI resolution hint forwarded by the `openai-responses` serializers (default stays `auto`) and by `openai-completions` for the values Chat Completions supports. `"original"` preserves native resolution — required for snapcompact frames, whose pixel-font glyphs do not survive the default downscale. Providers without a detail knob ignore the field.
-
-### Fixed
-
-- Fixed OpenRouter DeepSeek V4 strict tool schemas nesting `anyOf` inside the nullable wrapper for optional unions, which produced a branch without `type` and triggered OpenRouter's `Invalid tool parameters schema : field anyOf: missing field type` 400. ([#2270](https://github.com/can1357/oh-my-pi/issues/2270))
-- Hardened strict tool-schema handling beyond the optional-union case: `enforceStrictSchema` now splices natively nested pure unions into the parent `anyOf` (only when the inner node carries no constraining siblings, since sibling keywords are conjunctive with `anyOf`), so source schemas with nested unions no longer produce type-less `anyOf` branches that strict upstream validators reject. ([#2270](https://github.com/can1357/oh-my-pi/issues/2270))
-- Made the openai-completions non-strict retry reachable for `"mixed"` strict mode (previously gated to `all_strict`, i.e. Cerebras only) and taught it to recognize upstream tool-schema validation 400s (`Invalid tool parameters schema …`, `Invalid schema for function …`). A matching rejection now retries the request with base (non-strict) schemas and persists `strictToolsDisabled` on the provider session, so later requests skip the doomed strict attempt instead of paying a 400 + retry round-trip each turn. ([#2270](https://github.com/can1357/oh-my-pi/issues/2270))
-- Cross-model `anthropic-messages → anthropic-messages` continuations now preserve prior assistant turns' reasoning chains end-to-end: every prior `thinking`/`redactedThinking` block survives (not just the latest surviving assistant), and third-party ↔ third-party replays keep their signatures intact so the reasoning chain stays signed for the next turn. Signatures are stripped (and any `redacted_thinking` sibling without a native landing spot is dropped) only when an official Anthropic endpoint is on either end of the replay — official Anthropic cryptographically binds reasoning signatures to its key+session+model, while compatible reasoning endpoints (Z.AI, DeepSeek, custom anthropic-messages providers configured via `models.yaml`) treat them as opaque continuation hints. Source-side official detection uses the canonical catalog provider id `"anthropic"` (assistant messages carry no `baseUrl`); target-side detection reuses the baked `compat.officialEndpoint` flag. Latest-turn byte-for-byte behavior (Anthropic's "thinking blocks in the latest assistant message cannot be modified" rule) and existing aborted/errored last-block sanitization are unchanged. ([#2257](https://github.com/can1357/oh-my-pi/issues/2257), [#2265](https://github.com/can1357/oh-my-pi/issues/2265))
-
-Older entries are archived in [packages/ai/CHANGELOG.md@1f7329fc2c7c](https://github.com/can1357/oh-my-pi/blob/1f7329fc2c7c366b38731738e0db9c170f9bb348/packages/ai/CHANGELOG.md).
+Older entries are archived in [packages/ai/CHANGELOG.md@d58593a30902](https://github.com/can1357/oh-my-pi/blob/d58593a3090258473304608d68ffd1f620e6b695/packages/ai/CHANGELOG.md).

package/dist/types/usage/claude-reset.d.ts

@@ -25,9 +25,14 @@ export interface ClaudeResetConsumeResult {
     raw?: unknown;
 }
 /**
- * Reuse program blocks when a normal usage response actually includes them.
- * A lone non-redeemable Cedar block is inconclusive because Juniper requires
- * the separate at-wall read, so callers must still perform discovery.
+ * Reuse program blocks when a normal usage response actually evaluated them.
+ * Anthropic lists every promo program as a key on the plain `/usage` body and
+ * leaves the value `null` until the matching probe query (`?cedar_ember=1`,
+ * `?at_wall=1`) asks for an evaluation, so a `null` block is "unknown", never
+ * "no resets". Only an object block is an answer; anything else returns `null`
+ * so the caller falls through to {@link listClaudeResetCredits}. A lone
+ * non-redeemable Cedar block is likewise inconclusive because Juniper requires
+ * the separate at-wall read.
  *
  * @internal
  */

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-ai",
-	"version": "18.2.9",
+	"version": "18.2.11",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"keywords": [
 		"ai",
@@ -152,11 +152,11 @@
 		"fmt": "oxfmt --no-error-on-unmatched-pattern 'src/**/*.{ts,tsx}' '{test,bench,examples,scripts}/**/*.ts' '*.ts'"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.2.9",
-		"@oh-my-pi/pi-catalog": "18.2.9",
-		"@oh-my-pi/pi-natives": "18.2.9",
-		"@oh-my-pi/pi-utils": "18.2.9",
-		"@oh-my-pi/pi-wire": "18.2.9"
+		"@oh-my-pi/omptype": "18.2.11",
+		"@oh-my-pi/pi-catalog": "18.2.11",
+		"@oh-my-pi/pi-natives": "18.2.11",
+		"@oh-my-pi/pi-utils": "18.2.11",
+		"@oh-my-pi/pi-wire": "18.2.11"
 	},
 	"devDependencies": {
 		"@types/bun": "^1.3.14"

package/src/providers/anthropic.ts

@@ -466,8 +466,11 @@ type AnthropicControlState = {
 	stableSystemBlocks: AnthropicSystemBlock[] | undefined;
 	systemFingerprint: string | undefined;
 	controlTransitions: AnthropicControlTransition[];
-	baseEffort: AnthropicOutputEffort | undefined;
+	/** Whether the effort baseline was captured; `undefined` efforts are a valid baseline. */
+	effortBaselined: boolean;
+	/** Top-level `output_config.effort` of the baseline request; `undefined` = API default. */
 	baseEffortWire: AnthropicOutputEffort | undefined;
+	/** Effort in force at the conversation tail; `undefined` = API default. */
 	currentEffort: AnthropicOutputEffort | undefined;
 };
 
@@ -504,7 +507,7 @@ function createAnthropicControlState(): AnthropicControlState {
 		stableSystemBlocks: undefined,
 		systemFingerprint: undefined,
 		controlTransitions: [],
-		baseEffort: undefined,
+		effortBaselined: false,
 		baseEffortWire: undefined,
 		currentEffort: undefined,
 	};
@@ -3999,7 +4002,7 @@ function resetAnthropicControlState(state: AnthropicControlState): void {
 	state.stableSystemBlocks = undefined;
 	state.systemFingerprint = undefined;
 	state.controlTransitions = [];
-	state.baseEffort = undefined;
+	state.effortBaselined = false;
 	state.baseEffortWire = undefined;
 	state.currentEffort = undefined;
 }
@@ -4202,6 +4205,13 @@ function planStableAnthropicTools(
  * later changes as per-message effort. Anthropic applies a system message's
  * `output_config.effort` from the next `user` turn on, so the control is
  * anchored before the latest user message to take effect on this response.
+ *
+ * An omitted effort means the API's per-model default (`medium` on Opus 5.5,
+ * `high` elsewhere), so it is tracked as its own state rather than assumed to
+ * be any concrete level: every later explicit level is sent as a control. A
+ * per-message control cannot express "back to the API default", so a request
+ * that drops its effort mid-session keeps the level already in force instead
+ * of rewriting the top-level value and invalidating the cache.
  */
 function planStableAnthropicEffort(
 	current: AnthropicOutputEffort | undefined,
@@ -4210,18 +4220,17 @@ function planStableAnthropicEffort(
 	enabled: boolean,
 ): AnthropicOutputEffort | undefined {
 	if (!state || !enabled) return current;
-	const effective = current ?? "high";
-	if (state.baseEffort === undefined) {
-		state.baseEffort = effective;
+	if (!state.effortBaselined) {
+		state.effortBaselined = true;
 		state.baseEffortWire = current;
-		state.currentEffort = effective;
+		state.currentEffort = current;
 		return current;
 	}
-	if (state.currentEffort !== effective) {
+	if (current !== undefined && state.currentEffort !== current) {
 		const lastUserIndex = messages.findLastIndex(message => message.role === "user");
 		const messageCount = lastUserIndex >= 0 ? lastUserIndex : messages.length;
-		recordAnthropicControlTransition(state, messages, messageCount, [], effective);
-		state.currentEffort = effective;
+		recordAnthropicControlTransition(state, messages, messageCount, [], current);
+		state.currentEffort = current;
 	}
 	return state.baseEffortWire;
 }

package/src/usage/alibaba-token-plan.ts

@@ -93,9 +93,9 @@ function usageStatus(usedFraction: number): UsageLimit["status"] {
 }
 
 function buildLimit(
-	id: "5h" | "7d",
+	id: "5h" | "7d" | "monthly",
 	label: string,
-	durationMs: number,
+	durationMs: number | undefined,
 	usedFraction: number | undefined,
 	resetsAt: number | undefined,
 	accountId: string | undefined,
@@ -105,7 +105,7 @@ function buildLimit(
 		id: `credits:${id}`,
 		label,
 		scope: { provider: PROVIDER, ...(accountId ? { accountId } : {}), windowId: id },
-		window: { id, label, durationMs, ...(resetsAt ? { resetsAt } : {}) },
+		window: { id, label, ...(durationMs !== undefined ? { durationMs } : {}), ...(resetsAt ? { resetsAt } : {}) },
 		amount: { used: usedFraction * 100, usedFraction, unit: "percent" },
 		status: usageStatus(usedFraction),
 	};
@@ -245,6 +245,17 @@ async function fetchAlibabaTokenPlanUsage(
 				parsePositiveTimestamp(responseData.per1WeekResetTime),
 				accountId,
 			),
+			// Monthly-only plans report just this bucket, and the console never
+			// states its span (calendar months differ), so the window carries a
+			// reset deadline without a duration.
+			buildLimit(
+				"monthly",
+				"Monthly Credits",
+				undefined,
+				parseUsedFraction(responseData.per1MonthPercentage),
+				parsePositiveTimestamp(responseData.per1MonthResetTime),
+				accountId,
+			),
 		].filter((limit): limit is UsageLimit => limit !== undefined);
 		if (limits.length === 0) return null;
 		return {

package/src/usage/claude-reset.ts

@@ -448,9 +448,14 @@ async function withResolvedOrganization(
 }
 
 /**
- * Reuse program blocks when a normal usage response actually includes them.
- * A lone non-redeemable Cedar block is inconclusive because Juniper requires
- * the separate at-wall read, so callers must still perform discovery.
+ * Reuse program blocks when a normal usage response actually evaluated them.
+ * Anthropic lists every promo program as a key on the plain `/usage` body and
+ * leaves the value `null` until the matching probe query (`?cedar_ember=1`,
+ * `?at_wall=1`) asks for an evaluation, so a `null` block is "unknown", never
+ * "no resets". Only an object block is an answer; anything else returns `null`
+ * so the caller falls through to {@link listClaudeResetCredits}. A lone
+ * non-redeemable Cedar block is likewise inconclusive because Juniper requires
+ * the separate at-wall read.
  *
  * @internal
  */
@@ -459,26 +464,24 @@ export function parseClaudeResetCreditsFromUsagePayload(
 	orgId?: string,
 	baseUrl?: string,
 ): ClaudeResetCreditList | null {
-	if (!isRecord(payload) || !Object.hasOwn(payload, CEDAR_PROGRAM)) return null;
-	const parsedCedar = parseCedarStatus(payload[CEDAR_PROGRAM]);
-	if (!parsedCedar.ok) return null;
+	if (!isRecord(payload)) return null;
+	const cedarBlock = payload[CEDAR_PROGRAM];
+	const juniperBlock = payload[JUNIPER_PROGRAM];
 	const normalizedOrgId = orgId?.trim();
 	const safeOrgId = normalizedOrgId && ORGANIZATION_ID.test(normalizedOrgId) ? normalizedOrgId : undefined;
-	const cedarList = parsedCedar.status ? normalizeCedarList(parsedCedar.status, safeOrgId, baseUrl) : undefined;
-	if (cedarList && (cedarList.availableCount > 0 || cedarList.nextCreditId)) return cedarList;
-	if (!Object.hasOwn(payload, JUNIPER_PROGRAM)) return null;
-	const parsedJuniper = parseJuniperStatus(payload[JUNIPER_PROGRAM]);
-	if (!parsedJuniper.ok) return null;
-	if (parsedJuniper.status) return normalizeJuniperList(parsedJuniper.status, safeOrgId, baseUrl);
-	if (cedarList) return cedarList;
-	return {
-		availableCount: 0,
-		redeemableCount: 0,
-		eligible: false,
-		credits: [],
-		...(safeOrgId ? { orgId: safeOrgId } : {}),
-		...(baseUrl ? { baseUrl } : {}),
-	};
+	let cedarList: ClaudeResetCreditList | undefined;
+	if (isRecord(cedarBlock)) {
+		const parsedCedar = parseCedarStatus(cedarBlock);
+		if (!parsedCedar.ok || !parsedCedar.status) return null;
+		cedarList = normalizeCedarList(parsedCedar.status, safeOrgId, baseUrl);
+		if (cedarList.availableCount > 0 || cedarList.nextCreditId) return cedarList;
+	}
+	if (isRecord(juniperBlock)) {
+		const parsedJuniper = parseJuniperStatus(juniperBlock);
+		if (!parsedJuniper.ok || !parsedJuniper.status) return null;
+		return normalizeJuniperList(parsedJuniper.status, safeOrgId, baseUrl);
+	}
+	return cedarList ?? null;
 }
 
 /**