Merge pull request #5030 from usr-bin-roygbiv/fix/provider-native-compaction

fix(compaction): retry after AWS credential failures

packages/ai/CHANGELOG.md

@@ -22,6 +22,7 @@
 
 - Fixed concurrent reasoning summaries to ignore legacy streaming events under cutoff contract
 - Fixed sequential-cutoff Codex reasoning summaries repeating earlier content when atomic summary snapshots are replayed or extended.
+- Fixed error classification for typed AWS credential-resolution failures (`AwsCredentialsError`) to map them to authentication failures. ([#5030](https://github.com/can1357/oh-my-pi/pull/5030) by [@usr-bin-roygbiv](https://github.com/usr-bin-roygbiv))
 
 ## [16.3.15] - 2026-07-09
 

packages/ai/src/error/flags.ts

@@ -1,5 +1,6 @@
 import { isUnexpectedSocketCloseMessage } from "@oh-my-pi/pi-utils";
 import type { Api, AssistantMessage } from "../types";
+import { AwsCredentialsError } from "./aws";
 import {
 	AnthropicConnectionError,
 	AnthropicConnectionTimeoutError,
@@ -346,7 +347,9 @@ export function classify(error: unknown, api?: Api): number {
 			}
 		}
 
-		if (link instanceof AnthropicConnectionTimeoutError) {
+		if (link instanceof AwsCredentialsError) {
+			kinds |= Flag.AuthFailed;
+		} else if (link instanceof AnthropicConnectionTimeoutError) {
 			kinds |= Flag.Timeout | Flag.Transient;
 		} else if (link instanceof AnthropicConnectionError) {
 			kinds |= Flag.Transient;

packages/ai/test/error-aierr.test.ts

@@ -32,6 +32,11 @@ describe("AIError.classify — structural provider errors", () => {
 		).toBe(true);
 	});
 
+	it("classifies a typed AWS credential-resolution failure as authFailed", () => {
+		const id = AIError.classify(new AIError.AwsCredentialsError("opaque provider setup failure", "resolution"));
+		expect(AIError.is(id, AIError.Flag.AuthFailed)).toBe(true);
+	});
+
 	it("maps the usage_limit_reached code to usageLimit on a 429", () => {
 		const id = AIError.classify(
 			new AIError.ProviderHttpError("Payment Required", 429, { code: "usage_limit_reached" }),

packages/coding-agent/CHANGELOG.md

@@ -4,6 +4,7 @@
 
 ### Fixed
 
+- Fixed compaction aborting instead of trying an authenticated fallback model when Amazon Bedrock credential resolution fails before a request is sent. ([#5030](https://github.com/can1357/oh-my-pi/pull/5030) by [@usr-bin-roygbiv](https://github.com/usr-bin-roygbiv))
 - Fixed full-context forks cold-missing OpenAI prompt caches by persisting an inherited provider prompt-cache key separately from the new OMP session id, adding `--prompt-cache-key` for explicit cache affinity, and dropping automatic inheritance when startup changes the model, thinking level, system prompt, or tool schema. ([#5035](https://github.com/can1357/oh-my-pi/issues/5035))
 - Fixed Codex advisor requests using local `-advisor` session labels as provider session IDs; advisors now use stable UUIDv7 provider identities while keeping labeled transcript names. ([#5040](https://github.com/can1357/oh-my-pi/issues/5040))
 

packages/coding-agent/src/session/agent-session.ts

@@ -12024,18 +12024,6 @@ export class AgentSession {
 
 		return candidates;
 	}
-	#isCompactionAuthFailure(error: unknown): boolean {
-		if (!(error instanceof Error)) return false;
-		// Real provider 401/403 — surfaced as `.status` by the compaction layer
-		// (see `createSummarizationError` in packages/agent/src/compaction/compaction.ts).
-		// Without this branch, an expired/revoked Anthropic key would bypass the
-		// authenticated-fallback path and dump the raw HTTP body into the UI.
-		const status = (error as Error & { status?: number }).status;
-		if (status === 401 || status === 403) return true;
-		// pi-native gateway synthetic for "no credential configured" (issue #986).
-		// Carries no HTTP status, so the legacy message regex stays.
-		return /auth_unavailable|no auth available/i.test(error.message);
-	}
 
 	#buildCompactionAuthError(): Error {
 		const currentModel = this.model;
@@ -12101,7 +12089,7 @@ export class AgentSession {
 					},
 				);
 			} catch (error) {
-				if (!this.#isCompactionAuthFailure(error)) {
+				if (!AIError.is(AIError.classify(error, candidate.api), AIError.Flag.AuthFailed)) {
 					throw error;
 				}
 			}
@@ -12804,7 +12792,7 @@ export class AgentSession {
 
 							const message = error instanceof Error ? error.message : String(error);
 							const id = AIError.classify(error, candidate.api);
-							if (this.#isCompactionAuthFailure(error)) {
+							if (AIError.is(id, AIError.Flag.AuthFailed)) {
 								lastError = this.#buildCompactionAuthError();
 								break;
 							}

packages/coding-agent/test/compaction-prefer-current-model.test.ts

@@ -2,6 +2,7 @@ import { afterEach, beforeEach, describe, expect, it, vi } from "bun:test";
 import * as path from "node:path";
 import { Agent } from "@oh-my-pi/pi-agent-core";
 import * as compactionModule from "@oh-my-pi/pi-agent-core/compaction";
+import * as AIError from "@oh-my-pi/pi-ai/error";
 import { buildModel } from "@oh-my-pi/pi-catalog/build";
 import { getBundledModel } from "@oh-my-pi/pi-catalog/models";
 import { ModelRegistry } from "@oh-my-pi/pi-coding-agent/config/model-registry";
@@ -99,6 +100,76 @@ describe("compaction prefers the current session model over modelRoles.default",
 		expect(`${firstCandidate.provider}/${firstCandidate.id}`).toBe(`${currentModel.provider}/${currentModel.id}`);
 	});
 
+	it("falls back when the authenticated Bedrock candidate cannot resolve AWS credentials", async () => {
+		const currentModel = getBundledModel("amazon-bedrock", "global.anthropic.claude-opus-4-6-v1");
+		const fallbackModel = getBundledModel("anthropic", "claude-sonnet-4-5");
+		if (!currentModel || !fallbackModel) {
+			throw new Error("Expected bundled test models to exist");
+		}
+
+		const settings = Settings.isolated({ "compaction.keepRecentTokens": 1, "compaction.strategy": "context-full" });
+		settings.setModelRole("smol", `${fallbackModel.provider}/${fallbackModel.id}`);
+
+		const agent = new Agent({
+			initialState: {
+				model: currentModel,
+				systemPrompt: ["Test"],
+				tools: [],
+				messages: [],
+			},
+		});
+
+		authStorage = await AuthStorage.create(path.join(tempDir.path(), "testauth.db"));
+		authStorage.setRuntimeApiKey(currentModel.provider, "bedrock-credentials");
+		authStorage.setRuntimeApiKey(fallbackModel.provider, "anthropic-token");
+		modelRegistry = new ModelRegistry(authStorage, path.join(tempDir.path(), "models.yml"));
+
+		session = new AgentSession({
+			agent,
+			sessionManager: SessionManager.inMemory(),
+			settings,
+			modelRegistry,
+		});
+		session.subscribe(() => {});
+
+		for (const [userText, assistantText] of [
+			["first question", "first answer"],
+			["second question", "second answer"],
+		] as const) {
+			const user = userMsg(userText);
+			const assistant = assistantMsg(assistantText);
+			session.agent.appendMessage(user);
+			session.sessionManager.appendMessage(user);
+			session.agent.appendMessage(assistant);
+			session.sessionManager.appendMessage(assistant);
+		}
+
+		const compactSpy = vi.spyOn(compactionModule, "compact").mockImplementation(async (preparation, model) => {
+			if (model.provider === currentModel.provider && model.id === currentModel.id) {
+				throw new AIError.AwsCredentialsError("opaque provider setup failure", "resolution");
+			}
+			if (model.provider !== fallbackModel.provider || model.id !== fallbackModel.id) {
+				throw new Error(`Unexpected compaction model ${model.provider}/${model.id}`);
+			}
+			return {
+				summary: "fallback summary",
+				shortSummary: "fallback short summary",
+				firstKeptEntryId: preparation.firstKeptEntryId,
+				tokensBefore: 42,
+				details: { provider: model.provider },
+			};
+		});
+
+		const result = await session.compact();
+
+		expect(result.summary).toBe("fallback summary");
+		expect(compactSpy).toHaveBeenCalledTimes(2);
+		expect(compactSpy.mock.calls.map(([, model]) => `${model.provider}/${model.id}`)).toEqual([
+			`${currentModel.provider}/${currentModel.id}`,
+			`${fallbackModel.provider}/${fallbackModel.id}`,
+		]);
+	});
+
 	it("uses compactionModel only for the summary call and leaves the active model unchanged", async () => {
 		const baseCurrentModel = getBundledModel("anthropic", "claude-sonnet-4-5");
 		const compactionModel = getBundledModel("openai", "gpt-5");