pi-codebuddy-sdk 0.3.1 → 0.4.1

package/README.md

@@ -102,6 +102,25 @@ File: `~/.pi/agent/codebuddy-sdk.json` or `.pi/codebuddy-sdk.json` in a project.
 | `provider.appendSystemPrompt` | `true` | Use Pi's system prompt instead of CodeBuddy's default identity |
 | `provider.strictMcpConfig` | `true` | Use only Pi-bridged MCP tools |
 | `provider.pathToCodebuddyCode` | auto | Path to `codebuddy` when it is **not** on `PATH` |
+| `provider.contextWindow` | estimated | Override context window (tokens) for **all** models (e.g. `300000`) |
+| `provider.maxTokens` | estimated | Override max output tokens for **all** models (e.g. `32768`) |
+| `provider.modelOverrides` | — | Per-model overrides, keyed by model id or substring (see below) |
+
+`modelOverrides` lets you fine-tune individual models; the longest matching
+key wins and beats the global `contextWindow` / `maxTokens`:
+
+```json
+{
+  "provider": {
+    "modelOverrides": {
+      "gpt-5": { "contextWindow": 1048576, "maxTokens": 64000, "images": false },
+      "claude": { "reasoning": true }
+    }
+  }
+}
+```
+
+Supported per-model keys: `contextWindow`, `maxTokens`, `reasoning`, `images`.
 
 ## Privacy
 
@@ -117,6 +136,12 @@ export CODEBUDDY_SDK_DEBUG=1
 
 Default log: `~/.pi/agent/codebuddy-sdk.log`. See [CONTRIBUTING.md](CONTRIBUTING.md) for maintainer details.
 
+**After `pi -r`, the previously selected codebuddy model is gone from `/model`?**
+The last-known-good model list is cached to `~/.pi/agent/codebuddy-sdk-models.json` on every
+successful discovery, so a new pi process can re-register it synchronously at module load —
+even when SDK discovery is still in flight (new process, gate busy, or stale auth). If the
+model is still missing, share the debug log (`CODEBUDDY_SDK_DEBUG=1`) and the cache file.
+
 ## Development
 
 Maintainers: [CONTRIBUTING.md](CONTRIBUTING.md).

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-codebuddy-sdk",
-  "version": "0.3.1",
+  "version": "0.4.1",
   "description": "Pi extension that uses CodeBuddy (via Agent SDK) as a model provider and adds an AskCodebuddy tool.",
   "author": "slahser",
   "license": "MIT",

package/src/config.ts

@@ -27,6 +27,17 @@ export interface Config {
 		settingSources?: SettingSource[];
 		strictMcpConfig?: boolean;
 		pathToCodebuddyCode?: string;
+		/** Override the estimated context window (tokens) for ALL models. */
+		contextWindow?: number;
+		/** Override the estimated max output tokens for ALL models. */
+		maxTokens?: number;
+		/** Per-model overrides keyed by model id (or id substring). */
+		modelOverrides?: Record<string, {
+			contextWindow?: number;
+			maxTokens?: number;
+			reasoning?: boolean;
+			images?: boolean;
+		}>;
 	};
 }
 

package/src/index.ts

@@ -10,6 +10,7 @@ import { homedir } from "os";
 import { dirname, join } from "path";
 import { PROVIDER_ID, messageContentToText, convertPiMessages } from "./convert.js";
 import { buildModels, codebuddyModelId, FALLBACK_MODELS, rawModelsFromSdk, resolveModel as _resolveModel, type PiModel } from "./models.js";
+import { readModelsCache, writeModelsCache } from "./models-cache.js";
 import { MCP_SERVER_NAME, MCP_TOOL_PREFIX, buildCodebuddySystemPrompt } from "./skills.js";
 import { verifyWrittenSession as _verifyWrittenSession } from "./session-verify.js";
 import { extractAllToolResults as _extractAllToolResults, type McpResult } from "./extract-tool-results.js";
@@ -877,7 +878,18 @@ function processStreamEvent(
 	}
 
 	if (event?.type === "content_block_delta") {
-		const index = c.turnBlocks.findIndex((b: any) => b.index === event.index);
+		// Match the most-recently-started block at this content_block index.
+		// A later tool_use can reuse an index already held by an earlier
+		// block (seen in real streams); first-match lookup would append the
+		// later block's deltas onto the earlier one, concatenating two JSON
+		// payloads and silently dropping one tool's arguments. Reverse scan
+		// routes deltas to the active (last-started) block. (Stop events
+		// keep first-match: they run in start order and each deletes its
+		// index, so the next stop finds the correct block.)
+		let index = -1;
+		for (let i = c.turnBlocks.length - 1; i >= 0; i--) {
+			if (c.turnBlocks[i].index === event.index) { index = i; break; }
+		}
 		const block = c.turnBlocks[index];
 		if (!block) return;
 		if (event.delta?.type === "text_delta" && block.type === "text") {
@@ -1572,7 +1584,18 @@ async function discoverModels(pi: ExtensionAPI): Promise<void> {
 			const supported = await q.supportedModels();
 			await q.return().catch(() => {});
 			if (!supported.length) return;
-			MODELS = buildModels(rawModelsFromSdk(supported as any));
+			MODELS = buildModels(
+				rawModelsFromSdk(supported as any),
+				providerSettings.modelOverrides ? undefined : providerSettings,
+				providerSettings.modelOverrides,
+			);
+			// Persist the successful discovery globally so a later session
+			// resume (module reload) can recover the real model list even if
+			// discovery fails that time (e.g. SDK subprocess gate is busy).
+			// Written to disk so a NEW pi process (`pi -r` after exit) can also
+			// recover it synchronously at module load — the in-process global
+			// alone dies with the old process.
+			writeModelsCache(MODELS);
 			const g = globalThis as Record<symbol, any>;
 			const streamFn = g[ACTIVE_STREAM_SIMPLE_KEY] ?? streamCodebuddySdk;
 			pi.registerProvider(PROVIDER_ID, {
@@ -1586,7 +1609,29 @@ async function discoverModels(pi: ExtensionAPI): Promise<void> {
 			modelsDiscovered = true;
 			debug(`discoverModels: registered ${MODELS.length} models from SDK`);
 		} catch (err) {
-			debug("discoverModels: failed, using fallback models", err);
+			// Resume/reload: discovery may fail (SDK subprocess gate busy,
+			// stale auth, transient CLI error). Recover the last known-good
+			// model list instead of collapsing to the single fallback model —
+			// otherwise the previously selected codebuddy model disappears
+			// from /model after `pi -r` and the user must re-pick it.
+			const cached = readModelsCache();
+			if (cached && cached.length) {
+				MODELS = cached;
+				debug(`discoverModels: failed, recovered ${MODELS.length} cached models`, err);
+				const g = globalThis as Record<symbol, any>;
+				const streamFn = g[ACTIVE_STREAM_SIMPLE_KEY] ?? streamCodebuddySdk;
+				pi.registerProvider(PROVIDER_ID, {
+					name: "CodeBuddy",
+					baseUrl: PROVIDER_ID,
+					apiKey: "not-used",
+					api: "codebuddy-sdk",
+					models: MODELS as any,
+					streamSimple: streamFn as any,
+				});
+				modelsDiscovered = true;
+			} else {
+				debug("discoverModels: failed, using fallback models", err);
+			}
 		}
 	});
 }
@@ -1602,6 +1647,16 @@ export default function (pi: ExtensionAPI) {
 	const config = loadConfig(process.cwd());
 	debug("loadConfig:", JSON.stringify(config));
 	providerSettings = config.provider ?? {};
+	// Apply config overrides to the fallback models immediately, and recover
+	// a previously cached model list across module reloads (session resume)
+	// or across processes (`pi -r` after exit).
+	const cached = readModelsCache();
+	if (cached && cached.length) {
+		MODELS = cached;
+		debug(`default: recovered ${MODELS.length} cached models from prior discovery`);
+	} else {
+		MODELS = buildModels(FALLBACK_MODELS, providerSettings, providerSettings.modelOverrides);
+	}
 	const registeredModels = MODELS;
 
 	discoverInFlight = discoverModels(pi);

package/src/models-cache.ts

@@ -0,0 +1,48 @@
+// Persisted last-known-good model list.
+//
+// `pi -r` after exiting pi spawns a NEW process: the Symbol.for in-process
+// cache from v0.4.0 dies with the old process, so a session-resume model
+// resolution that runs while SDK discovery is still in flight only sees the
+// fallback list and the previously selected codebuddy model drops out of
+// /model (user must re-pick it). Persisting the list to disk lets the new
+// process register the full model list synchronously at module load.
+
+import { homedir } from "os";
+import { dirname, join } from "path";
+import { mkdirSync, readFileSync, renameSync, writeFileSync } from "fs";
+import type { PiModel } from "./models.js";
+
+export const MODELS_CACHE_KEY = Symbol.for("codebuddy-sdk:modelsCache");
+
+const defaultCacheFile = () => join(homedir(), ".pi", "agent", "codebuddy-sdk-models.json");
+
+// Injectable for tests.
+const cacheFile = () => process.env.CODEBUDDY_SDK_MODELS_CACHE_PATH || defaultCacheFile();
+
+export function readModelsCache(): PiModel[] | undefined {
+	// In-process shared cache first (module reloads within a process).
+	const inProcess = (globalThis as Record<symbol, unknown>)[MODELS_CACHE_KEY] as PiModel[] | undefined;
+	if (inProcess?.length) return inProcess;
+	// Disk cache second (new process — `pi -r` after exiting pi).
+	try {
+		const parsed: unknown = JSON.parse(readFileSync(cacheFile(), "utf8"));
+		if (Array.isArray(parsed) && parsed.length) return parsed as PiModel[];
+	} catch {
+		// No cache yet, or corrupt file — caller falls back.
+	}
+	return undefined;
+}
+
+export function writeModelsCache(models: PiModel[]): void {
+	if (!models.length) return;
+	(globalThis as Record<symbol, unknown>)[MODELS_CACHE_KEY] = models;
+	try {
+		const path = cacheFile();
+		mkdirSync(dirname(path), { recursive: true });
+		const tmp = `${path}.tmp`;
+		writeFileSync(tmp, JSON.stringify(models));
+		renameSync(tmp, path);
+	} catch {
+		// Cache is best-effort; discovery retries next load.
+	}
+}

package/src/models.ts

@@ -55,11 +55,49 @@ export const FALLBACK_MODELS: PiModel[] = [
 	{ id: "hy3-preview-agent-ioa", name: "Hunyuan 3 Preview", reasoning: true, input: ["text"], contextWindow: DEFAULT_CONTEXT, maxTokens: DEFAULT_MAX_TOKENS, cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 } },
 ];
 
-export function buildModels(models: PiModel[]): PiModel[] {
-	return models.map((m) => ({
-		...m,
-		cost: m.cost ?? { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
-	}));
+export type ModelOverrides = {
+	contextWindow?: number;
+	maxTokens?: number;
+	reasoning?: boolean;
+	images?: boolean;
+};
+
+/**
+ * Apply config-driven overrides on top of the estimated model metadata.
+ * `globalOverrides` applies to every model; `perModel` is keyed by model id
+ * (matched by exact id first, then case-insensitive substring, longest key
+ * wins) and beats the global defaults.
+ */
+export function buildModels(
+	models: PiModel[],
+	globalOverrides?: ModelOverrides,
+	perModel?: Record<string, ModelOverrides>,
+): PiModel[] {
+	const sortedKeys = Object.keys(perModel ?? {}).sort((a, b) => b.length - a.length);
+	const match = (id: string): ModelOverrides | undefined => {
+		const lower = id.toLowerCase();
+		for (const key of sortedKeys) {
+			const k = key.toLowerCase();
+			if (id === key || (lower.includes(k) && k.length > 0)) return perModel![key];
+		}
+		return undefined;
+	};
+	return models.map((m) => {
+		const o = { ...globalOverrides, ...match(m.id) };
+		const contextWindow = o.contextWindow ?? m.contextWindow;
+		const maxTokens = o.maxTokens ?? m.maxTokens;
+		const input: ("text" | "image")[] = o.images === undefined
+			? m.input
+			: (o.images ? ["text", "image"] as const : ["text"] as const);
+		return {
+			...m,
+			contextWindow,
+			maxTokens,
+			reasoning: o.reasoning ?? m.reasoning,
+			input,
+			cost: m.cost ?? { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+		};
+	});
 }
 
 export function codebuddyModelId(model: { id: string }): string {