add branch/compact retries to agent-harness
This commit is contained in:
@@ -4,6 +4,8 @@ import {
|
||||
type ImageContent,
|
||||
type Model,
|
||||
type Models,
|
||||
type RetryCallbacks,
|
||||
type RetryPolicy,
|
||||
type UserMessage,
|
||||
} from "@earendil-works/pi-ai";
|
||||
import { runAgentLoop } from "../agent-loop.ts";
|
||||
@@ -178,6 +180,7 @@ export class AgentHarness<
|
||||
private thinkingLevel: ThinkingLevel;
|
||||
private systemPrompt: AgentHarnessOptions<TSkill, TPromptTemplate, TTool>["systemPrompt"];
|
||||
private streamOptions: AgentHarnessStreamOptions;
|
||||
private retry: RetryPolicy | undefined;
|
||||
private resources: AgentHarnessResources<TSkill, TPromptTemplate>;
|
||||
private tools = new Map<string, TTool>();
|
||||
private activeToolNames: string[];
|
||||
@@ -194,6 +197,7 @@ export class AgentHarness<
|
||||
this.models = options.models;
|
||||
this.resources = options.resources ?? {};
|
||||
this.streamOptions = cloneStreamOptions(options.streamOptions);
|
||||
this.retry = options.retry;
|
||||
this.systemPrompt = options.systemPrompt;
|
||||
this.validateUniqueNames(
|
||||
(options.tools ?? []).map((tool) => tool.name),
|
||||
@@ -256,6 +260,15 @@ export class AgentHarness<
|
||||
return lastResult;
|
||||
}
|
||||
|
||||
private retryCallbacks(operation: "compaction" | "branch_summary"): RetryCallbacks {
|
||||
return {
|
||||
onRetryScheduled: (attempt, maxAttempts, delayMs, errorMessage) =>
|
||||
this.emitOwn({ type: "retry_scheduled", operation, attempt, maxAttempts, delayMs, errorMessage }),
|
||||
onRetryAttemptStart: () => this.emitOwn({ type: "retry_attempt_start", operation }),
|
||||
onRetryFinished: () => this.emitOwn({ type: "retry_finished", operation }),
|
||||
};
|
||||
}
|
||||
|
||||
private async emitBeforeProviderRequest(
|
||||
model: Model<any>,
|
||||
sessionId: string,
|
||||
@@ -720,7 +733,16 @@ export class AgentHarness<
|
||||
const provided = hookResult?.compaction;
|
||||
const compactResult = provided
|
||||
? { ok: true as const, value: provided }
|
||||
: await compact(preparation, this.models, model, customInstructions, undefined, this.thinkingLevel);
|
||||
: await compact(
|
||||
preparation,
|
||||
this.models,
|
||||
model,
|
||||
customInstructions,
|
||||
undefined,
|
||||
this.thinkingLevel,
|
||||
this.retry,
|
||||
this.retryCallbacks("compaction"),
|
||||
);
|
||||
if (!compactResult.ok) throw compactResult.error;
|
||||
const result = compactResult.value;
|
||||
const entryId = await this.session.appendCompaction(
|
||||
@@ -781,6 +803,8 @@ export class AgentHarness<
|
||||
signal: new AbortController().signal,
|
||||
customInstructions: hookResult?.customInstructions ?? options?.customInstructions,
|
||||
replaceInstructions: hookResult?.replaceInstructions ?? options?.replaceInstructions,
|
||||
retry: this.retry,
|
||||
callbacks: this.retryCallbacks("branch_summary"),
|
||||
});
|
||||
if (!branchSummary.ok) {
|
||||
if (branchSummary.error.code === "aborted") return { cancelled: true };
|
||||
|
||||
@@ -1,4 +1,4 @@
|
||||
import { contentText, type Model, type Models } from "@earendil-works/pi-ai";
|
||||
import { contentText, type Model, type Models, type RetryCallbacks, type RetryPolicy } from "@earendil-works/pi-ai";
|
||||
|
||||
import type { AgentMessage } from "../../types.ts";
|
||||
import {
|
||||
@@ -9,7 +9,7 @@ import {
|
||||
} from "../messages.ts";
|
||||
import type { BranchSummaryResult, Session, SessionTreeEntry } from "../types.ts";
|
||||
import { BranchSummaryError, err, ok, type Result, SessionError } from "../types.ts";
|
||||
import { estimateTokens, SUMMARIZATION_SYSTEM_PROMPT } from "./compaction.ts";
|
||||
import { completeSimpleWithRetries, estimateTokens, SUMMARIZATION_SYSTEM_PROMPT } from "./compaction.ts";
|
||||
import {
|
||||
computeFileLists,
|
||||
createFileOps,
|
||||
@@ -61,6 +61,10 @@ export interface GenerateBranchSummaryOptions {
|
||||
replaceInstructions?: boolean;
|
||||
/** Tokens reserved for prompt and model output. Defaults to 16384. */
|
||||
reserveTokens?: number;
|
||||
/** Optional retry policy for transient summarization errors. */
|
||||
retry?: RetryPolicy;
|
||||
/** Optional callbacks for retry reporting. */
|
||||
callbacks?: RetryCallbacks;
|
||||
}
|
||||
|
||||
/** Collect entries that should be summarized before navigating to a different session tree entry. */
|
||||
@@ -200,7 +204,16 @@ export async function generateBranchSummary(
|
||||
entries: SessionTreeEntry[],
|
||||
options: GenerateBranchSummaryOptions,
|
||||
): Promise<Result<BranchSummaryResult, BranchSummaryError>> {
|
||||
const { models, model, signal, customInstructions, replaceInstructions, reserveTokens = 16384 } = options;
|
||||
const {
|
||||
models,
|
||||
model,
|
||||
signal,
|
||||
customInstructions,
|
||||
replaceInstructions,
|
||||
reserveTokens = 16384,
|
||||
retry,
|
||||
callbacks,
|
||||
} = options;
|
||||
const contextWindow = model.contextWindow || 128000;
|
||||
const tokenBudget = contextWindow - reserveTokens;
|
||||
|
||||
@@ -228,10 +241,13 @@ export async function generateBranchSummary(
|
||||
timestamp: Date.now(),
|
||||
},
|
||||
];
|
||||
const response = await models.completeSimple(
|
||||
const response = await completeSimpleWithRetries(
|
||||
models,
|
||||
model,
|
||||
{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },
|
||||
{ signal, maxTokens: 2048 },
|
||||
retry,
|
||||
callbacks,
|
||||
);
|
||||
if (response.stopReason === "aborted") {
|
||||
return err(new BranchSummaryError("aborted", response.errorMessage || "Branch summary aborted"));
|
||||
|
||||
@@ -1,9 +1,14 @@
|
||||
import {
|
||||
type AssistantMessage,
|
||||
type Context,
|
||||
contentText,
|
||||
type ImageContent,
|
||||
type Model,
|
||||
type Models,
|
||||
type RetryCallbacks,
|
||||
type RetryPolicy,
|
||||
retryAssistantCall,
|
||||
type SimpleStreamOptions,
|
||||
type TextContent,
|
||||
type Usage,
|
||||
} from "@earendil-works/pi-ai";
|
||||
@@ -107,6 +112,17 @@ export interface CompactionResult<T = unknown> {
|
||||
details?: T;
|
||||
}
|
||||
|
||||
export async function completeSimpleWithRetries(
|
||||
models: Models,
|
||||
model: Model<any>,
|
||||
context: Context,
|
||||
options: SimpleStreamOptions,
|
||||
retry?: RetryPolicy,
|
||||
callbacks?: RetryCallbacks,
|
||||
): Promise<AssistantMessage> {
|
||||
return retryAssistantCall(() => models.completeSimple(model, context, options), retry, options.signal, callbacks);
|
||||
}
|
||||
|
||||
function combineUsage(first: Usage, second: Usage): Usage {
|
||||
return {
|
||||
input: first.input + second.input,
|
||||
@@ -499,6 +515,8 @@ export async function generateSummary(
|
||||
customInstructions?: string,
|
||||
previousSummary?: string,
|
||||
thinkingLevel?: ThinkingLevel,
|
||||
retry?: RetryPolicy,
|
||||
callbacks?: RetryCallbacks,
|
||||
): Promise<Result<string, CompactionError>> {
|
||||
const result = await generateSummaryWithUsage(
|
||||
currentMessages,
|
||||
@@ -509,6 +527,8 @@ export async function generateSummary(
|
||||
customInstructions,
|
||||
previousSummary,
|
||||
thinkingLevel,
|
||||
retry,
|
||||
callbacks,
|
||||
);
|
||||
return result.ok ? ok(result.value.text) : err(result.error);
|
||||
}
|
||||
@@ -523,6 +543,8 @@ export async function generateSummaryWithUsage(
|
||||
customInstructions?: string,
|
||||
previousSummary?: string,
|
||||
thinkingLevel?: ThinkingLevel,
|
||||
retry?: RetryPolicy,
|
||||
callbacks?: RetryCallbacks,
|
||||
): Promise<Result<{ text: string; usage: Usage }, CompactionError>> {
|
||||
const maxTokens = Math.min(
|
||||
Math.floor(0.8 * reserveTokens),
|
||||
@@ -553,10 +575,13 @@ export async function generateSummaryWithUsage(
|
||||
? { maxTokens, signal, reasoning: thinkingLevel }
|
||||
: { maxTokens, signal };
|
||||
|
||||
const response = await models.completeSimple(
|
||||
const response = await completeSimpleWithRetries(
|
||||
models,
|
||||
model,
|
||||
{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },
|
||||
completionOptions,
|
||||
retry,
|
||||
callbacks,
|
||||
);
|
||||
if (response.stopReason === "aborted") {
|
||||
return err(new CompactionError("aborted", response.errorMessage || "Summarization aborted"));
|
||||
@@ -688,6 +713,8 @@ export async function compact(
|
||||
customInstructions?: string,
|
||||
signal?: AbortSignal,
|
||||
thinkingLevel?: ThinkingLevel,
|
||||
retry?: RetryPolicy,
|
||||
callbacks?: RetryCallbacks,
|
||||
): Promise<Result<CompactionResult, CompactionError>> {
|
||||
const {
|
||||
firstKeptEntryId,
|
||||
@@ -720,6 +747,8 @@ export async function compact(
|
||||
customInstructions,
|
||||
previousSummary,
|
||||
thinkingLevel,
|
||||
retry,
|
||||
callbacks,
|
||||
);
|
||||
if (!historyResult.ok) return err(historyResult.error);
|
||||
historyText = historyResult.value.text;
|
||||
@@ -732,6 +761,8 @@ export async function compact(
|
||||
settings.reserveTokens,
|
||||
signal,
|
||||
thinkingLevel,
|
||||
retry,
|
||||
callbacks,
|
||||
);
|
||||
if (!turnPrefixResult.ok) return err(turnPrefixResult.error);
|
||||
summary = `${historyText}\n\n---\n\n**Turn Context (split turn):**\n\n${turnPrefixResult.value.text}`;
|
||||
@@ -748,6 +779,8 @@ export async function compact(
|
||||
customInstructions,
|
||||
previousSummary,
|
||||
thinkingLevel,
|
||||
retry,
|
||||
callbacks,
|
||||
);
|
||||
if (!summaryResult.ok) return err(summaryResult.error);
|
||||
summary = summaryResult.value.text;
|
||||
@@ -772,6 +805,8 @@ async function generateTurnPrefixSummary(
|
||||
reserveTokens: number,
|
||||
signal?: AbortSignal,
|
||||
thinkingLevel?: ThinkingLevel,
|
||||
retry?: RetryPolicy,
|
||||
callbacks?: RetryCallbacks,
|
||||
): Promise<Result<{ text: string; usage: Usage }, CompactionError>> {
|
||||
const maxTokens = Math.min(
|
||||
Math.floor(0.5 * reserveTokens),
|
||||
@@ -788,12 +823,17 @@ async function generateTurnPrefixSummary(
|
||||
},
|
||||
];
|
||||
|
||||
const response = await models.completeSimple(
|
||||
model,
|
||||
{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },
|
||||
const completionOptions =
|
||||
model.reasoning && thinkingLevel && thinkingLevel !== "off"
|
||||
? { maxTokens, signal, reasoning: thinkingLevel }
|
||||
: { maxTokens, signal },
|
||||
: { maxTokens, signal };
|
||||
const response = await completeSimpleWithRetries(
|
||||
models,
|
||||
model,
|
||||
{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },
|
||||
completionOptions,
|
||||
retry,
|
||||
callbacks,
|
||||
);
|
||||
if (response.stopReason === "aborted") {
|
||||
return err(new CompactionError("aborted", response.errorMessage || "Turn prefix summarization aborted"));
|
||||
|
||||
@@ -2,6 +2,7 @@ import type {
|
||||
ImageContent,
|
||||
Model,
|
||||
Models,
|
||||
RetryPolicy,
|
||||
SimpleStreamOptions,
|
||||
TextContent,
|
||||
Transport,
|
||||
@@ -613,6 +614,25 @@ export interface SessionTreeEvent {
|
||||
fromHook?: boolean;
|
||||
}
|
||||
|
||||
export interface RetryScheduledEvent {
|
||||
type: "retry_scheduled";
|
||||
operation: "compaction" | "branch_summary";
|
||||
attempt: number;
|
||||
maxAttempts: number;
|
||||
delayMs: number;
|
||||
errorMessage: string;
|
||||
}
|
||||
|
||||
export interface RetryAttemptStartEvent {
|
||||
type: "retry_attempt_start";
|
||||
operation: "compaction" | "branch_summary";
|
||||
}
|
||||
|
||||
export interface RetryFinishedEvent {
|
||||
type: "retry_finished";
|
||||
operation: "compaction" | "branch_summary";
|
||||
}
|
||||
|
||||
export interface ModelUpdateEvent {
|
||||
type: "model_update";
|
||||
model: Model<any>;
|
||||
@@ -663,6 +683,9 @@ export type AgentHarnessOwnEvent<
|
||||
| SessionCompactEvent
|
||||
| SessionBeforeTreeEvent
|
||||
| SessionTreeEvent
|
||||
| RetryScheduledEvent
|
||||
| RetryAttemptStartEvent
|
||||
| RetryFinishedEvent
|
||||
| ModelUpdateEvent
|
||||
| ThinkingLevelUpdateEvent
|
||||
| ResourcesUpdateEvent<TSkill, TPromptTemplate>
|
||||
@@ -732,6 +755,9 @@ export type AgentHarnessEventResultMap = {
|
||||
session_compact: undefined;
|
||||
session_before_tree: SessionBeforeTreeResult | undefined;
|
||||
session_tree: undefined;
|
||||
retry_scheduled: undefined;
|
||||
retry_attempt_start: undefined;
|
||||
retry_finished: undefined;
|
||||
model_update: undefined;
|
||||
thinking_level_update: undefined;
|
||||
resources_update: undefined;
|
||||
@@ -848,6 +874,8 @@ export interface AgentHarnessOptions<
|
||||
}) => string | Promise<string>);
|
||||
/** Curated stream/provider request options. Snapshotted at turn start. */
|
||||
streamOptions?: AgentHarnessStreamOptions;
|
||||
/** Optional retry policy for generated compaction and branch-summary requests. */
|
||||
retry?: RetryPolicy;
|
||||
model: Model<any>;
|
||||
thinkingLevel?: ThinkingLevel;
|
||||
activeToolNames?: string[];
|
||||
|
||||
@@ -550,6 +550,180 @@ describe("AgentHarness", () => {
|
||||
expect(compaction?.type === "compaction" ? compaction.usage : undefined).toEqual(usage);
|
||||
});
|
||||
|
||||
describe("summarization retries", () => {
|
||||
it("retries transient compaction errors and emits retry events", async () => {
|
||||
const registration = newFaux();
|
||||
let calls = 0;
|
||||
registration.setResponses([
|
||||
() => {
|
||||
calls++;
|
||||
return fauxAssistantMessage("", { stopReason: "error", errorMessage: "terminated" });
|
||||
},
|
||||
() => {
|
||||
calls++;
|
||||
return fauxAssistantMessage("## Goal\nRecovered summary");
|
||||
},
|
||||
]);
|
||||
const session = new Session(new InMemorySessionStorage());
|
||||
await session.appendMessage(createUserMessage("one"));
|
||||
await session.appendMessage(createAssistantMessage("two"));
|
||||
const harness = new AgentHarness({
|
||||
models,
|
||||
env: new NodeExecutionEnv({ cwd: process.cwd() }),
|
||||
session,
|
||||
model: registration.getModel(),
|
||||
retry: { enabled: true, maxRetries: 1, baseDelayMs: 0 },
|
||||
});
|
||||
const retryEvents: string[] = [];
|
||||
harness.subscribe((event) => {
|
||||
if (
|
||||
event.type === "retry_scheduled" ||
|
||||
event.type === "retry_attempt_start" ||
|
||||
event.type === "retry_finished"
|
||||
) {
|
||||
retryEvents.push(`${event.type}:${event.operation}`);
|
||||
}
|
||||
});
|
||||
|
||||
const result = await harness.compact();
|
||||
|
||||
expect(result.summary).toContain("Recovered summary");
|
||||
expect(calls).toBe(2);
|
||||
expect(retryEvents).toEqual([
|
||||
"retry_scheduled:compaction",
|
||||
"retry_attempt_start:compaction",
|
||||
"retry_finished:compaction",
|
||||
]);
|
||||
});
|
||||
|
||||
it("does not retry non-retryable compaction errors", async () => {
|
||||
const registration = newFaux();
|
||||
let calls = 0;
|
||||
registration.setResponses([
|
||||
() => {
|
||||
calls++;
|
||||
return fauxAssistantMessage("", { stopReason: "error", errorMessage: "insufficient_quota" });
|
||||
},
|
||||
]);
|
||||
const session = new Session(new InMemorySessionStorage());
|
||||
await session.appendMessage(createUserMessage("one"));
|
||||
await session.appendMessage(createAssistantMessage("two"));
|
||||
const harness = new AgentHarness({
|
||||
models,
|
||||
env: new NodeExecutionEnv({ cwd: process.cwd() }),
|
||||
session,
|
||||
model: registration.getModel(),
|
||||
retry: { enabled: true, maxRetries: 1, baseDelayMs: 0 },
|
||||
});
|
||||
const retryEvents: string[] = [];
|
||||
harness.subscribe((event) => {
|
||||
if (
|
||||
event.type === "retry_scheduled" ||
|
||||
event.type === "retry_attempt_start" ||
|
||||
event.type === "retry_finished"
|
||||
) {
|
||||
retryEvents.push(event.type);
|
||||
}
|
||||
});
|
||||
|
||||
await expect(harness.compact()).rejects.toThrow("insufficient_quota");
|
||||
|
||||
expect(calls).toBe(1);
|
||||
expect(retryEvents).toEqual([]);
|
||||
});
|
||||
|
||||
it("exhausts transient compaction retries after maxRetries failures", async () => {
|
||||
const registration = newFaux();
|
||||
let calls = 0;
|
||||
registration.setResponses(
|
||||
Array.from({ length: 4 }, () => () => {
|
||||
calls++;
|
||||
return fauxAssistantMessage("", { stopReason: "error", errorMessage: "terminated" });
|
||||
}),
|
||||
);
|
||||
const session = new Session(new InMemorySessionStorage());
|
||||
await session.appendMessage(createUserMessage("one"));
|
||||
await session.appendMessage(createAssistantMessage("two"));
|
||||
const harness = new AgentHarness({
|
||||
models,
|
||||
env: new NodeExecutionEnv({ cwd: process.cwd() }),
|
||||
session,
|
||||
model: registration.getModel(),
|
||||
retry: { enabled: true, maxRetries: 3, baseDelayMs: 0 },
|
||||
});
|
||||
const retryEvents: string[] = [];
|
||||
harness.subscribe((event) => {
|
||||
if (
|
||||
event.type === "retry_scheduled" ||
|
||||
event.type === "retry_attempt_start" ||
|
||||
event.type === "retry_finished"
|
||||
) {
|
||||
retryEvents.push(`${event.type}:${event.operation}`);
|
||||
}
|
||||
});
|
||||
|
||||
await expect(harness.compact()).rejects.toThrow("terminated");
|
||||
|
||||
expect(calls).toBe(4);
|
||||
expect(retryEvents).toEqual([
|
||||
"retry_scheduled:compaction",
|
||||
"retry_attempt_start:compaction",
|
||||
"retry_scheduled:compaction",
|
||||
"retry_attempt_start:compaction",
|
||||
"retry_scheduled:compaction",
|
||||
"retry_attempt_start:compaction",
|
||||
"retry_finished:compaction",
|
||||
]);
|
||||
});
|
||||
|
||||
it("retries transient branch summary errors and emits retry events", async () => {
|
||||
const registration = newFaux();
|
||||
let calls = 0;
|
||||
registration.setResponses([
|
||||
() => {
|
||||
calls++;
|
||||
return fauxAssistantMessage("", { stopReason: "error", errorMessage: "terminated" });
|
||||
},
|
||||
() => {
|
||||
calls++;
|
||||
return fauxAssistantMessage("## Goal\nRecovered branch summary");
|
||||
},
|
||||
]);
|
||||
const session = new Session(new InMemorySessionStorage());
|
||||
const targetId = await session.appendMessage(createUserMessage("first branch"));
|
||||
await session.appendMessage(createAssistantMessage("first reply"));
|
||||
await session.appendMessage(createUserMessage("abandoned work"));
|
||||
await session.appendMessage(createAssistantMessage("abandoned reply"));
|
||||
const harness = new AgentHarness({
|
||||
models,
|
||||
env: new NodeExecutionEnv({ cwd: process.cwd() }),
|
||||
session,
|
||||
model: registration.getModel(),
|
||||
retry: { enabled: true, maxRetries: 1, baseDelayMs: 0 },
|
||||
});
|
||||
const retryEvents: string[] = [];
|
||||
harness.subscribe((event) => {
|
||||
if (
|
||||
event.type === "retry_scheduled" ||
|
||||
event.type === "retry_attempt_start" ||
|
||||
event.type === "retry_finished"
|
||||
) {
|
||||
retryEvents.push(`${event.type}:${event.operation}`);
|
||||
}
|
||||
});
|
||||
|
||||
const result = await harness.navigateTree(targetId, { summarize: true });
|
||||
|
||||
expect(result.summaryEntry?.summary).toContain("Recovered branch summary");
|
||||
expect(calls).toBe(2);
|
||||
expect(retryEvents).toEqual([
|
||||
"retry_scheduled:branch_summary",
|
||||
"retry_attempt_start:branch_summary",
|
||||
"retry_finished:branch_summary",
|
||||
]);
|
||||
});
|
||||
});
|
||||
|
||||
it("persists generated branch summary usage", async () => {
|
||||
const registration = newFaux();
|
||||
registration.setResponses([fauxAssistantMessage("## Goal\nBranch summary")]);
|
||||
|
||||
Reference in New Issue
Block a user