@oh-my-pi/pi-catalog 16.2.12 → 16.2.13

package/CHANGELOG.md

@@ -2,6 +2,16 @@
 
 ## [Unreleased]
 
+## [16.2.13] - 2026-07-01
+
+### Added
+
+- Added support for human-readable reasoning summaries on compatible OpenAI Codex models (v5.4+)
+
+### Fixed
+
+- Fixed discovered OpenAI Codex models to advertise V2 streaming remote compaction, avoiding the legacy compact endpoint timeout path for Codex sessions. ([#4146](https://github.com/can1357/oh-my-pi/issues/4146))
+
 ## [16.2.12] - 2026-07-01
 
 ### Breaking Changes

package/dist/types/identity/family.d.ts

@@ -60,6 +60,15 @@ export declare const isOpenAIModelId: (modelId: string) => boolean;
  * fall back to omitting `context`, letting the server default to `current_turn`.
  */
 export declare const supportsAllTurnsReasoningContext: (modelId: string) => boolean;
+/**
+ * OpenAI Codex models that accept `reasoning.summary`. Shares the gpt-5.4 wire
+ * floor with {@link supportsAllTurnsReasoningContext}: earlier Codex ids
+ * (`gpt-5.1-codex`, `gpt-5.3-codex`, `gpt-5.3-codex-spark`) reject the field
+ * with `Unsupported parameter: 'reasoning.summary' is not supported with this
+ * model`. Callers omit `summary` for unsupported ids, letting the server skip
+ * the human-readable summary stream.
+ */
+export declare const supportsCodexReasoningSummary: (modelId: string) => boolean;
 /**
  * Reasoning-capable GLM coding SKUs: glm-4.5 and up on the base / `-air` /
  * `-turbo` lines. Excludes the vision (`…v`) shape, the non-reasoning

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "16.2.12",
+	"version": "16.2.13",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -34,12 +34,12 @@
 	},
 	"dependencies": {
 		"@bufbuild/protobuf": "^2.12.0",
-		"@oh-my-pi/pi-utils": "16.2.12",
+		"@oh-my-pi/pi-utils": "16.2.13",
 		"arktype": "^2.2.0",
 		"zod": "^4"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "16.2.12",
+		"@oh-my-pi/pi-ai": "16.2.13",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/discovery/codex.ts

@@ -8,6 +8,11 @@ const DEFAULT_CONTEXT_WINDOW = 272_000;
 const DEFAULT_MAX_TOKENS = 128_000;
 const DEFAULT_CODEX_CLIENT_VERSION = "0.99.0";
 const NPM_CODEX_LATEST_URL = "https://registry.npmjs.org/@openai%2Fcodex/latest";
+const CODEX_REMOTE_COMPACTION = {
+	enabled: true,
+	api: "openai-codex-responses",
+	v2StreamingEnabled: true,
+} as const;
 
 const codexReasoningPresetSchema = type({
 	"effort?": "unknown",
@@ -269,6 +274,7 @@ function normalizeCodexModelEntry(entry: unknown, baseUrl: string): NormalizedCo
 			reasoning,
 			input,
 			cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+			remoteCompaction: CODEX_REMOTE_COMPACTION,
 			contextWindow,
 			maxTokens,
 			...(preferWebsockets ? { preferWebsockets: true } : {}),

package/src/identity/family.ts

@@ -123,6 +123,13 @@ export const isOpenAIModelId = memo((modelId: string): boolean => {
 	return /(^|\/)(gpt|o1|o3|o4)[-.]/i.test(modelId) || modelId.toLowerCase().includes("openai/");
 });
 
+/** OpenAI models at or above the gpt-5.4 wire generation, keyed off the parsed version. */
+const isOpenAIWireGen54Plus = memo((modelId: string): boolean => {
+	const parsed = parseOpenAIModel(bareModelId(modelId));
+	if (!parsed) return false;
+	return semverGte(parsed.version, "5.4");
+});
+
 /**
  * OpenAI Codex models that honor `reasoning.context: "all_turns"` (full
  * cross-turn reasoning replay). The `reasoning.context` field itself exists for
@@ -133,11 +140,17 @@ export const isOpenAIModelId = memo((modelId: string): boolean => {
  * floor (not an allowlist) so 5.6/6.x inherit support automatically. Callers
  * fall back to omitting `context`, letting the server default to `current_turn`.
  */
-export const supportsAllTurnsReasoningContext = memo((modelId: string): boolean => {
-	const parsed = parseOpenAIModel(bareModelId(modelId));
-	if (!parsed) return false;
-	return semverGte(parsed.version, "5.4");
-});
+export const supportsAllTurnsReasoningContext = isOpenAIWireGen54Plus;
+
+/**
+ * OpenAI Codex models that accept `reasoning.summary`. Shares the gpt-5.4 wire
+ * floor with {@link supportsAllTurnsReasoningContext}: earlier Codex ids
+ * (`gpt-5.1-codex`, `gpt-5.3-codex`, `gpt-5.3-codex-spark`) reject the field
+ * with `Unsupported parameter: 'reasoning.summary' is not supported with this
+ * model`. Callers omit `summary` for unsupported ids, letting the server skip
+ * the human-readable summary stream.
+ */
+export const supportsCodexReasoningSummary = isOpenAIWireGen54Plus;
 
 /**
  * Reasoning-capable GLM coding SKUs: glm-4.5 and up on the base / `-air` /

package/src/model-cache.ts

@@ -7,14 +7,15 @@ import { getModelDbPath } from "@oh-my-pi/pi-utils";
 import type { Api, Model, ModelSpec } from "./types";
 
 // Rows persist ModelSpec JSON (sparse `compat`, never the resolved record);
-// the model manager rebuilds via `buildModel` on load. v7 invalidates rows
-// predating the Antigravity Gemini budget-mode migration (cached specs still
-// carrying `thinking.mode: "google-level"` and the old 3.5-flash effort
-// routing); v6 invalidates rows that may contain the retired unknown-limit
-// sentinels (222222/8888); v5 invalidated rows predating effort-tier variant
-// collapsing (raw `-low`/`-high`/`-thinking` member ids); v4 dropped the
-// pre-efforts ThinkingConfig shape.
-const CACHE_SCHEMA_VERSION = 7;
+// the model manager rebuilds via `buildModel` on load. v8 invalidates Codex
+// discovery rows predating provider-native V2 compaction metadata; v7
+// invalidated rows predating the Antigravity Gemini budget-mode migration
+// (cached specs still carrying `thinking.mode: "google-level"` and the old
+// 3.5-flash effort routing); v6 invalidated rows that may contain the retired
+// unknown-limit sentinels (222222/8888); v5 invalidated rows predating
+// effort-tier variant collapsing (raw `-low`/`-high`/`-thinking` member ids);
+// v4 dropped the pre-efforts ThinkingConfig shape.
+const CACHE_SCHEMA_VERSION = 8;
 
 interface CacheRow {
 	provider_id: string;
@@ -100,7 +101,12 @@ function migrateCacheSchema(db: Database): void {
 	} finally {
 		stmt.finalize();
 	}
-	db.run("UPDATE model_cache SET version = ? WHERE version = 2", [CACHE_SCHEMA_VERSION]);
+	// Delete rows written under any older schema so they cannot be reused. The
+	// legacy `UPDATE ... WHERE version = 2` migration silently promoted the very
+	// first cache version to whatever the current one is, defeating every
+	// subsequent invalidation (see #4146: pre-V2 Codex rows kept the legacy
+	// compaction path even after CACHE_SCHEMA_VERSION was bumped).
+	db.run("DELETE FROM model_cache WHERE version <> ?", [CACHE_SCHEMA_VERSION]);
 }
 
 export function readModelCache<TApi extends Api>(