fix(runtime): compact context inside worker

This commit is contained in:
ZacharyZhang-NY committed 2026-04-22 14:19:46 -04:00
1 parent 33c6ede381
commit 6dfa1a851d
13 files changed
+214 -92

No files matched your search

+37 -19
View File
@@ -12,10 +12,9 @@
* `buildAgentHooks` collapses the wiring into a single object whose
* shape mirrors the subset of `AgentOptions` that we populate. The
* factory spreads the result into `new Agent({...})`, so the agent
* contract is still the pi-defined one — this is a pure organizational
* refactor. Behavior (RPC shapes, failure modes, tool-name semantics)
* is preserved verbatim by routing through the same `WorkerRpcClient`
* the factory previously held.
* contract is still the pi-defined one. Tool policy stays routed
* through `WorkerRpcClient`; context compaction runs inside the worker
* so the full transcript never crosses the worker-main boundary.
*
* The signatures below are pinned to pi via `pi-hook-contract`: any
* drift in `AgentOptions.beforeToolCall` / `afterToolCall` /
@@ -34,21 +33,28 @@ import type {
AfterToolCallRpcResult,
BeforeToolCallRpcPayload,
BeforeToolCallRpcResult,
TransformContextRpcPayload,
TransformContextRpcResult,
} from "./hook-rpc";
import type { WorkerRpcClient } from "../worker/entry/rpc";
import {
compactMessagesIfNeeded,
type CompactionTelemetry,
} from "./context-compactor";
export interface ContextCompactionTelemetry extends CompactionTelemetry {
runId: string;
emittedAt: string;
}
/**
* Inputs the aggregator needs to produce a hook bundle. `rpc` is the
* worker-side RPC client that ships hook invocations back to main
* where the actual policy (approval gate, artifact projection, context
* compaction) lives. `runId` is the routing key main uses to look up
* the per-run `ActiveRunContext` in `RuntimeBridge`.
* worker-side RPC client that ships policy hook invocations back to
* main. `runId` is the routing key main uses to look up the per-run
* `ActiveRunContext` in `RuntimeBridge`.
*/
export interface AgentHooksDeps {
runId: string;
rpc: WorkerRpcClient;
onContextCompaction?: (telemetry: ContextCompactionTelemetry) => void;
}
/**
@@ -74,11 +80,11 @@ export interface AgentHookBundle {
* must never silently bypass main's approval gate.
* - `afterToolCall`: fail-open RPC. Projection is observational; if
* main cannot be reached the tool result stands unchanged.
* - `transformContext`: fail-open RPC. The pi contract forbids throws
* here, so on RPC failure we return the original transcript.
* - `transformContext`: worker-local compaction. The transcript stays
* inside the worker; main receives telemetry only.
*/
export function buildAgentHooks(deps: AgentHooksDeps): AgentHookBundle {
const { runId, rpc } = deps;
const { runId, rpc, onContextCompaction } = deps;
const beforeToolCall: AgentHookBundle["beforeToolCall"] = async (
ctx,
@@ -129,12 +135,24 @@ export function buildAgentHooks(deps: AgentHooksDeps): AgentHookBundle {
const transformContext: AgentHookBundle["transformContext"] = async (
messages,
): Promise<AgentMessage[]> => {
const payload: TransformContextRpcPayload = { messages };
const response = await rpc.invoke(runId, "transformContext", payload);
if (!response.ok) return messages;
const result = response.result as TransformContextRpcResult;
if (result.changed && result.messages) return result.messages;
return messages;
try {
const outcome = compactMessagesIfNeeded(messages);
if (outcome.changed && outcome.messages) {
if (outcome.telemetry) {
onContextCompaction?.({
...outcome.telemetry,
runId,
emittedAt: new Date().toISOString(),
});
}
return outcome.messages;
}
return messages;
} catch (error) {
throw new Error(
`transformContext compaction failed: ${error instanceof Error ? error.message : String(error)}`,
);
}
};
return { beforeToolCall, afterToolCall, transformContext };
+24 -4
View File
@@ -51,6 +51,15 @@ interface MessageLike {
export interface CompactionOutcome {
changed: boolean;
messages: AgentMessage[] | null;
telemetry: CompactionTelemetry | null;
}
export interface CompactionTelemetry {
estimatedTokens: number;
originalMessageCount: number;
compactedMessageCount: number;
compactedOlderMessageCount: number;
retainedRecentMessageCount: number;
}
/**
@@ -69,16 +78,16 @@ export interface CompactionOutcome {
export function compactMessagesIfNeeded(messages: AgentMessage[]): CompactionOutcome {
const estimatedTokens = estimateTokens(messages);
if (estimatedTokens < COMPACTION_TRIGGER_TOKENS) {
return { changed: false, messages: null };
return { changed: false, messages: null, telemetry: null };
}
if (messages.length <= RECENT_MESSAGE_TAIL + 1) {
return { changed: false, messages: null };
return { changed: false, messages: null, telemetry: null };
}
const desiredCutoff = messages.length - RECENT_MESSAGE_TAIL;
const cutoff = findSafeCutoff(messages, desiredCutoff);
if (cutoff === null) {
return { changed: false, messages: null };
return { changed: false, messages: null, telemetry: null };
}
const older = messages.slice(0, cutoff);
@@ -91,7 +100,18 @@ export function compactMessagesIfNeeded(messages: AgentMessage[]): CompactionOut
timestamp: Date.now(),
} as AgentMessage;
return { changed: true, messages: [summary, ...recent] };
const compactedMessages = [summary, ...recent];
return {
changed: true,
messages: compactedMessages,
telemetry: {
estimatedTokens,
originalMessageCount: messages.length,
compactedMessageCount: compactedMessages.length,
compactedOlderMessageCount: older.length,
retainedRecentMessageCount: recent.length,
},
};
}
/**
+2 -22
View File
@@ -1,15 +1,13 @@
/**
* Main-side implementations of pi's tool-execution hooks. Invoked by
* `RuntimeBridge.handleRpc` when the worker proxies `beforeToolCall`,
* `afterToolCall`, or `transformContext`.
* `RuntimeBridge.handleRpc` when the worker proxies `beforeToolCall`
* or `afterToolCall`.
*
* Each handler is deliberately isolated to one concern:
* - `beforeToolCall`: operational-approval gate (blocks destructive
* tools only while an approval that gates EXECUTION is pending)
* - `afterToolCall`: filesystem artifact projection (bash/write/edit
* outputs become `work_products`)
* - `transformContext`: per-LLM-call auto-compaction via a pure
* reducer
*
* Policy lives here; `RuntimeBridge` stays a pure IPC router.
*/
@@ -20,14 +18,11 @@ import type {
AfterToolCallRpcResult,
BeforeToolCallRpcPayload,
BeforeToolCallRpcResult,
TransformContextRpcPayload,
TransformContextRpcResult,
} from "./hook-rpc";
import {
projectToolExecutionToWorkProduct,
type ArtifactProjectorRunContext,
} from "./artifact-projector";
import { compactMessagesIfNeeded } from "./context-compactor";
import {
isCofounderMandatoryApproval,
isOperationalApproval,
@@ -194,18 +189,3 @@ export function handleAfterToolCall(
}
return NO_OP_AFTER;
}
/**
* Context-transform hook. Delegates to `compactMessagesIfNeeded`,
* which is a pure, provider-free deterministic compactor. Runs before
* every LLM call; short transcripts short-circuit on the token-budget
* check so the steady-state cost is negligible.
*/
export function handleTransformContext(
_deps: HookHandlerDeps,
_ctx: ActiveRunContext,
payload: TransformContextRpcPayload,
): TransformContextRpcResult {
const outcome = compactMessagesIfNeeded(payload.messages);
return { changed: outcome.changed, messages: outcome.messages };
}
+6 -24
View File
@@ -1,21 +1,19 @@
/**
* Hook RPC contracts — the parent↔worker half of pi-agent-core's
* `beforeToolCall` / `afterToolCall` / `transformContext` hooks.
* `beforeToolCall` / `afterToolCall` hooks.
*
* The Agent instance lives in the utility process. Policy (approval,
* compaction, filesystem projection) lives in main. The hooks are
* therefore proxied: the worker invokes an RPC that main answers from
* the same `RuntimeBridge` that already owns agent-event and
* `persistMessage` handling.
* filesystem projection) lives in main. The hooks are therefore proxied:
* the worker invokes an RPC that main answers from the same
* `RuntimeBridge` that already owns agent-event and `persistMessage`
* handling. Context compaction runs in the worker process and reports
* telemetry through a worker message.
*
* Payloads intentionally exclude pi's `AgentContext` (systemPrompt,
* messages, tools) because main policy never needs it. Stripping it
* keeps each hook IPC well under the structured-clone budget even for
* large transcripts.
*/
import type { AgentMessage } from "@mariozechner/pi-agent-core";
// ─── beforeToolCall ───────────────────────────────────────────────────────
export interface BeforeToolCallRpcPayload {
@@ -51,19 +49,3 @@ export interface AfterToolCallRpcResult {
contentOverride: { type: "text"; text: string }[] | null;
isErrorOverride: boolean | null;
}
// ─── transformContext ─────────────────────────────────────────────────────
export interface TransformContextRpcPayload {
messages: AgentMessage[];
}
export interface TransformContextRpcResult {
/**
* Transformed transcript. When `changed` is false the worker keeps its
* original array instance instead of replacing it — saves structured
* clone on the common "no compaction yet" path.
*/
changed: boolean;
messages: AgentMessage[] | null;
}
+3 -18
View File
@@ -20,14 +20,12 @@ import { captureBaselineIfRequired, reconcileWorkspaceArtifacts } from "./artifa
import {
handleBeforeToolCall,
handleAfterToolCall,
handleTransformContext,
type ActiveRunContext,
type ActiveRunLifecycleState,
} from "./hook-handlers";
import type {
AfterToolCallRpcPayload,
BeforeToolCallRpcPayload,
TransformContextRpcPayload,
} from "./hook-rpc";
import { routeInternalToolCall } from "./internal-tools/router";
import type { InvokeInternalToolRpcPayload } from "@main/worker/runtime/internal-tools/internal-tool-rpc";
@@ -85,9 +83,9 @@ type RpcMethod = WorkerRpcMethod;
* - Forwards `agent-event` from the worker to the renderer (push channel).
* - Persists each `message_end` message via `SqliteSessionStore`.
* - Calls `CostWriter` for assistant messages with usage.
* - Proxies pi's tool-execution hooks (`beforeToolCall` / `afterToolCall`
* / `transformContext`) into policy handlers: approval-gate checks,
* filesystem artifact projection, and context compaction.
* - Proxies pi's tool-execution hooks (`beforeToolCall` / `afterToolCall`)
* into policy handlers: approval-gate checks and filesystem artifact
* projection.
*
* Policy itself lives in `hook-handlers.ts`; this class is a pure IPC
* router plus an `ActiveRunContext` cache.
@@ -563,8 +561,6 @@ export class RuntimeBridge {
return this.handleBeforeToolCallRpc(request.runId, request.payload as BeforeToolCallRpcPayload);
case "afterToolCall":
return this.handleAfterToolCallRpc(request.runId, request.payload as AfterToolCallRpcPayload);
case "transformContext":
return this.handleTransformContextRpc(request.runId, request.payload as TransformContextRpcPayload);
case "invokeInternalTool":
return this.handleInvokeInternalToolRpc(request.runId, request.payload as InvokeInternalToolRpcPayload);
default: {
@@ -733,15 +729,4 @@ export class RuntimeBridge {
return { ok: true, result };
}
private handleTransformContextRpc(
runId: string,
payload: TransformContextRpcPayload,
): { ok: true; result: unknown } | { ok: false; error: string } {
const ctx = this.activeRuns.get(runId);
if (!ctx) {
return { ok: false, error: `transformContext: no active run ${runId}` };
}
const result = handleTransformContext(this.deps, ctx, payload);
return { ok: true, result };
}
}
+4
View File
@@ -3,6 +3,7 @@ import type { workerMessageSchema } from "./types";
import { handleRunLog } from "../handlers/run-log-handler";
import { handleRunStatus } from "../handlers/run-status-handler";
import { handleRunFinished } from "../handlers/run-finished-handler";
import { handleContextCompaction } from "../handlers/context-compaction-handler";
import type { HandlerContext, HandlerDeps } from "../handlers/handler-context";
export async function handleWorkerMessage(
@@ -67,6 +68,9 @@ export async function handleWorkerMessage(
}
return;
}
case "context-compaction":
handleContextCompaction(event, deps, context);
return;
case "rpc-request": {
const bridge = deps.getRuntimeBridge?.();
const response = bridge
+10 -1
View File
@@ -24,7 +24,6 @@ export const workerRpcMethodSchema = z.enum([
"prepareRun",
"beforeToolCall",
"afterToolCall",
"transformContext",
"invokeInternalTool",
]);
@@ -65,6 +64,16 @@ export const workerMessageSchema = z.union([
emittedAt: z.string(),
event: z.record(z.string(), z.unknown()).and(z.object({ type: z.string() })),
}),
z.object({
type: z.literal("context-compaction"),
runId: z.string().uuid(),
emittedAt: z.string(),
estimatedTokens: z.number(),
originalMessageCount: z.number().int(),
compactedMessageCount: z.number().int(),
compactedOlderMessageCount: z.number().int(),
retainedRecentMessageCount: z.number().int(),
}),
z.object({
type: z.literal("rpc-request"),
rpcId: z.string(),
@@ -0,0 +1,30 @@
import { persistRunEvent } from "../observability/run-event-recorder";
import type { HandlerContext, HandlerDeps, WorkerMessageEvent } from "./handler-context";
export function handleContextCompaction(
event: Extract<WorkerMessageEvent, { type: "context-compaction" }>,
deps: HandlerDeps,
_context: HandlerContext,
): void {
const runState = deps.db.runs.getRunAuthState(event.runId);
if (!runState) {
deps.logger.warn(`[context-compaction] run ${event.runId} disappeared before telemetry could be recorded.`);
return;
}
persistRunEvent(deps.db, deps.logger, {
companyId: runState.companyId,
runId: event.runId,
agentId: runState.agentId,
eventType: "context_compaction",
level: "info",
message: `Compacted ${event.compactedOlderMessageCount} older messages into a summary.`,
payload: {
emittedAt: event.emittedAt,
estimatedTokens: event.estimatedTokens,
originalMessageCount: event.originalMessageCount,
compactedMessageCount: event.compactedMessageCount,
compactedOlderMessageCount: event.compactedOlderMessageCount,
retainedRecentMessageCount: event.retainedRecentMessageCount,
},
});
}
+10
View File
@@ -77,6 +77,16 @@ export type ChildMessage =
emittedAt: string;
event: { type: string } & Record<string, unknown>;
}
| {
type: "context-compaction";
runId: string;
emittedAt: string;
estimatedTokens: number;
originalMessageCount: number;
compactedMessageCount: number;
compactedOlderMessageCount: number;
retainedRecentMessageCount: number;
}
| { type: "rpc-request"; rpcId: string; runId: string; method: string; payload: unknown }
| { type: "run-agent-state"; runId: string; state: RunAgentLifecycleState }
| { type: "steer-ack"; runId: string; correlationId: string; accepted: boolean };
+12
View File
@@ -178,6 +178,18 @@ export class WorkerRunner {
apiKey: prepared.apiKey,
extraHeaders,
rpc,
onContextCompaction: (telemetry) => {
post({
type: "context-compaction",
runId: telemetry.runId,
emittedAt: telemetry.emittedAt,
estimatedTokens: telemetry.estimatedTokens,
originalMessageCount: telemetry.originalMessageCount,
compactedMessageCount: telemetry.compactedMessageCount,
compactedOlderMessageCount: telemetry.compactedOlderMessageCount,
retainedRecentMessageCount: telemetry.retainedRecentMessageCount,
});
},
});
this.activeAgents.set(job.runId, agent);
post({ type: "run-agent-state", runId: job.runId, state: "agent_active" });
@@ -9,7 +9,7 @@
* none of which the worker should replicate.
*
* This is structurally the same pattern as `beforeToolCall`/
* `afterToolCall`/`transformContext` in `hook-rpc.ts`; what's different
* `afterToolCall` in `hook-rpc.ts`; what's different
* is that those are side-effect hooks around pi's own tools, whereas
* this is a primary-effect tool whose business logic lives entirely in
* main. Keeping the payload shape minimal (name + JSON args) mirrors
+9 -3
View File
@@ -9,8 +9,8 @@
* thrown hooks — a thrown `beforeToolCall` produces an error tool
* result, a thrown `afterToolCall` likewise — so RPC failures degrade
* to a deterministic tool-error instead of crashing the run. The
* `transformContext` contract explicitly forbids throwing, so on RPC
* failure we return the original messages unchanged.
* `transformContext` runs the deterministic compactor in the worker
* process and lets compaction failures fail the run explicitly.
*/
import {
@@ -20,6 +20,7 @@ import {
} from "@mariozechner/pi-agent-core";
import { streamSimple, type Api, type Model } from "@mariozechner/pi-ai";
import { buildAgentHooks } from "@main/agent-runtime/agent-hooks";
import type { ContextCompactionTelemetry } from "@main/agent-runtime/agent-hooks";
import type { WorkerRpcClient } from "../entry/rpc";
// Side-effect import: drags the hook signature contract into the typecheck
// graph so any drift in pi-agent-core hook shapes breaks the build here.
@@ -40,6 +41,7 @@ export interface PiAgentFactoryInput {
apiKey: string;
extraHeaders: Record<string, string> | undefined;
rpc: WorkerRpcClient;
onContextCompaction?: (telemetry: ContextCompactionTelemetry) => void;
}
/** Map the UI-facing `thinkingEffort` string onto pi's `ThinkingLevel`. */
@@ -61,7 +63,11 @@ export function resolveThinkingLevel(effort: string | null | undefined): Thinkin
}
export function createPiAgent(input: PiAgentFactoryInput): Agent {
const hooks = buildAgentHooks({ runId: input.runId, rpc: input.rpc });
const hooks = buildAgentHooks({
runId: input.runId,
rpc: input.rpc,
onContextCompaction: input.onContextCompaction,
});
return new Agent({
initialState: {
systemPrompt: input.systemPrompt,
@@ -0,0 +1,66 @@
import type { AgentMessage } from "@mariozechner/pi-agent-core";
import { describe, expect, it, vi } from "vitest";
import { buildAgentHooks } from "../../src/main/agent-runtime/agent-hooks";
import type { WorkerRpcClient } from "../../src/main/worker/entry/rpc";
function createRpc() {
return {
invoke: vi.fn(async () => ({ ok: true, result: undefined })),
} as unknown as WorkerRpcClient & { invoke: ReturnType<typeof vi.fn> };
}
function largeTranscript(): AgentMessage[] {
const messages: AgentMessage[] = [];
for (let i = 0; i < 46; i += 1) {
const role = i % 2 === 0 ? "user" : "assistant";
messages.push({
role,
content: [{
type: "text",
text: `${role} turn ${i}\n${"context ".repeat(6_000)}`,
}],
timestamp: i,
} as AgentMessage);
}
return messages;
}
describe("worker-side context compaction", () => {
it("compacts inside the worker without transformContext RPC", async () => {
const rpc = createRpc();
const onContextCompaction = vi.fn();
const hooks = buildAgentHooks({
runId: "8b3f07b2-841a-4e16-9d1c-9d2a27c3c2ad",
rpc,
onContextCompaction,
});
const messages = largeTranscript();
const transformed = await hooks.transformContext(messages);
expect(rpc.invoke).not.toHaveBeenCalledWith(
expect.any(String),
"transformContext",
expect.anything(),
);
expect(transformed.length).toBeLessThan(messages.length);
expect(transformed[0]).toMatchObject({ role: "user" });
expect(onContextCompaction).toHaveBeenCalledWith(expect.objectContaining({
runId: "8b3f07b2-841a-4e16-9d1c-9d2a27c3c2ad",
originalMessageCount: messages.length,
compactedMessageCount: transformed.length,
}));
});
it("propagates compaction failures instead of returning the original transcript", async () => {
const rpc = createRpc();
const hooks = buildAgentHooks({
runId: "b9f6aa93-0db7-4f39-8f28-2d87742ef414",
rpc,
});
await expect(hooks.transformContext(null as unknown as AgentMessage[])).rejects.toThrow(
"transformContext compaction failed",
);
});
});