fix: carry observed overflow token counts into compaction (#40357)

Merged via squash. Prepared head SHA: b99eed4329 Co-authored-by: rabsef-bicrym <52549148+rabsef-bicrym@users.noreply.github.com> Co-authored-by: jalehman <550978+jalehman@users.noreply.github.com> Reviewed-by: @jalehman
2026-03-12 06:58:42 -07:00 · 2026-03-12 06:58:42 -07:00 · ff47876e61
parent f2e28fc30f
commit ff47876e61
28 changed files with 218 additions and 61 deletions
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@ -236,6 +236,7 @@ Docs: https://docs.openclaw.ai
 - Memory/Gemini: normalize returned Gemini embeddings across direct query, direct batch, and async batch paths so memory search uses consistent vector handling for Gemini too. (#43409) Thanks @gumadeiras.
 - Agents/failover: recognize additional serialized network errno strings plus `EHOSTDOWN` and `EPIPE` structured codes so transient transport failures trigger timeout failover more reliably. (#42830) Thanks @jnMetaCode.
 - Telegram/model picker: make inline model button selections persist the chosen session model correctly, clear overrides when selecting the configured default, and include effective fallback models in `/models` button validation. (#40105) Thanks @avirweb.
+- Agents/embedded runner: carry provider-observed overflow token counts into compaction so overflow retries and diagnostics use the rejected live prompt size instead of only transcript estimates. (#40357) thanks @rabsef-bicrym.

 ## 2026.3.7

--- a/src/agents/anthropic-payload-log.test.ts
+++ b/src/agents/anthropic-payload-log.test.ts
@ -29,7 +29,7 @@ describe("createAnthropicPayloadLogger", () => {
      ],
    };
    const streamFn: StreamFn = ((model, __, options) => {
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      return {} as never;
    }) as StreamFn;

--- a/src/agents/anthropic-payload-log.ts
+++ b/src/agents/anthropic-payload-log.ts
@ -145,7 +145,7 @@ export function createAnthropicPayloadLogger(params: {
          payload: redactedPayload,
          payloadDigest: digest(redactedPayload),
        });
-        return options?.onPayload?.(payload, model);
+        return options?.onPayload?.(payload);
      };
      return streamFn(model, context, {
        ...options,
--- a/src/agents/auth-profiles/oauth.openai-codex-refresh-fallback.test.ts
+++ b/src/agents/auth-profiles/oauth.openai-codex-refresh-fallback.test.ts
@ -17,7 +17,7 @@ const { getOAuthApiKeyMock } = vi.hoisted(() => ({
  }),
 }));

-vi.mock("@mariozechner/pi-ai/oauth", () => ({
+vi.mock("@mariozechner/pi-ai", () => ({
  getOAuthApiKey: getOAuthApiKeyMock,
  getOAuthProviders: () => [
    { id: "openai-codex", envApiKey: "OPENAI_API_KEY", oauthTokenEnv: "OPENAI_OAUTH_TOKEN" }, // pragma: allowlist secret
--- a/src/agents/auth-profiles/oauth.ts
+++ b/src/agents/auth-profiles/oauth.ts
@ -1,5 +1,9 @@
-import type { OAuthCredentials, OAuthProvider } from "@mariozechner/pi-ai";
-import { getOAuthApiKey, getOAuthProviders } from "@mariozechner/pi-ai/oauth";
+import {
+  getOAuthApiKey,
+  getOAuthProviders,
+  type OAuthCredentials,
+  type OAuthProvider,
+} from "@mariozechner/pi-ai";
 import { loadConfig, type OpenClawConfig } from "../../config/config.js";
 import { coerceSecretRef } from "../../config/types.secrets.js";
 import { withFileLock } from "../../infra/file-lock.js";
--- a/src/agents/openai-ws-stream.ts
+++ b/src/agents/openai-ws-stream.ts
@ -797,7 +797,7 @@ export function createOpenAIWebSocketStreamFn(
        ...(prevResponseId ? { previous_response_id: prevResponseId } : {}),
        ...extraParams,
      };
-      const nextPayload = await options?.onPayload?.(payload, model);
+      const nextPayload = options?.onPayload?.(payload);
      const requestPayload = (nextPayload ?? payload) as Parameters<
        OpenAIWebSocketManager["send"]
      >[0];
--- a/src/agents/pi-embedded-helpers.isbillingerrormessage.test.ts
+++ b/src/agents/pi-embedded-helpers.isbillingerrormessage.test.ts
@ -2,6 +2,7 @@ import { describe, expect, it } from "vitest";
 import {
  classifyFailoverReason,
  classifyFailoverReasonFromHttpStatus,
+  extractObservedOverflowTokenCount,
  isAuthErrorMessage,
  isAuthPermanentErrorMessage,
  isBillingErrorMessage,
@ -461,6 +462,29 @@ describe("isLikelyContextOverflowError", () => {
  });
 });

+describe("extractObservedOverflowTokenCount", () => {
+  it("extracts provider-reported prompt token counts", () => {
+    expect(
+      extractObservedOverflowTokenCount(
+        '400 {"type":"error","error":{"message":"prompt is too long: 277403 tokens > 200000 maximum"}}',
+      ),
+    ).toBe(277403);
+    expect(
+      extractObservedOverflowTokenCount("Context window exceeded: requested 12000 tokens"),
+    ).toBe(12000);
+    expect(
+      extractObservedOverflowTokenCount(
+        "This model's maximum context length is 128000 tokens. However, your messages resulted in 145000 tokens.",
+      ),
+    ).toBe(145000);
+  });
+
+  it("returns undefined when overflow counts are not present", () => {
+    expect(extractObservedOverflowTokenCount("Prompt too large for this model")).toBeUndefined();
+    expect(extractObservedOverflowTokenCount("rate limit exceeded")).toBeUndefined();
+  });
+});
+
 describe("isTransientHttpError", () => {
  it("returns true for retryable 5xx status codes", () => {
    expect(isTransientHttpError("499 Client Closed Request")).toBe(true);
--- a/src/agents/pi-embedded-helpers.ts
+++ b/src/agents/pi-embedded-helpers.ts
@ -22,6 +22,7 @@ export {
  isAuthPermanentErrorMessage,
  isModelNotFoundErrorMessage,
  isBillingAssistantError,
+  extractObservedOverflowTokenCount,
  parseApiErrorInfo,
  sanitizeUserFacingText,
  isBillingErrorMessage,
--- a/src/agents/pi-embedded-helpers/errors.ts
+++ b/src/agents/pi-embedded-helpers/errors.ts
@ -185,6 +185,32 @@ export function isCompactionFailureError(errorMessage?: string): boolean {
  return lower.includes("context overflow");
 }

+const OBSERVED_OVERFLOW_TOKEN_PATTERNS = [
+  /prompt is too long:\s*([\d,]+)\s+tokens\s*>\s*[\d,]+\s+maximum/i,
+  /requested\s+([\d,]+)\s+tokens/i,
+  /resulted in\s+([\d,]+)\s+tokens/i,
+];
+
+export function extractObservedOverflowTokenCount(errorMessage?: string): number | undefined {
+  if (!errorMessage) {
+    return undefined;
+  }
+
+  for (const pattern of OBSERVED_OVERFLOW_TOKEN_PATTERNS) {
+    const match = errorMessage.match(pattern);
+    const rawCount = match?.[1]?.replaceAll(",", "");
+    if (!rawCount) {
+      continue;
+    }
+    const parsed = Number(rawCount);
+    if (Number.isFinite(parsed) && parsed > 0) {
+      return Math.floor(parsed);
+    }
+  }
+
+  return undefined;
+}
+
 const ERROR_PAYLOAD_PREFIX_RE =
  /^(?:error|api\s*error|apierror|openai\s*error|anthropic\s*error|gateway\s*error)[:\s-]+/i;
 const FINAL_TAG_RE = /<\s*\/?\s*final\s*>/gi;
--- a/src/agents/pi-embedded-runner-extraparams.test.ts
+++ b/src/agents/pi-embedded-runner-extraparams.test.ts
@ -208,7 +208,7 @@ describe("applyExtraParamsToAgent", () => {
  }) {
    const payload = params.payload ?? { store: false };
    const baseStreamFn: StreamFn = (model, _context, options) => {
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      return {} as ReturnType<StreamFn>;
    };
    const agent = { streamFn: baseStreamFn };
@ -233,7 +233,7 @@ describe("applyExtraParamsToAgent", () => {
  }) {
    const payload = params.payload ?? {};
    const baseStreamFn: StreamFn = (model, _context, options) => {
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      return {} as ReturnType<StreamFn>;
    };
    const agent = { streamFn: baseStreamFn };
@ -276,7 +276,7 @@ describe("applyExtraParamsToAgent", () => {
    const payloads: Record<string, unknown>[] = [];
    const baseStreamFn: StreamFn = (_model, _context, options) => {
      const payload: Record<string, unknown> = { model: "deepseek/deepseek-r1" };
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -308,7 +308,7 @@ describe("applyExtraParamsToAgent", () => {
    const payloads: Record<string, unknown>[] = [];
    const baseStreamFn: StreamFn = (_model, _context, options) => {
      const payload: Record<string, unknown> = {};
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -332,7 +332,7 @@ describe("applyExtraParamsToAgent", () => {
    const payloads: Record<string, unknown>[] = [];
    const baseStreamFn: StreamFn = (_model, _context, options) => {
      const payload: Record<string, unknown> = { reasoning_effort: "high" };
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -357,7 +357,7 @@ describe("applyExtraParamsToAgent", () => {
    const payloads: Record<string, unknown>[] = [];
    const baseStreamFn: StreamFn = (_model, _context, options) => {
      const payload: Record<string, unknown> = { reasoning: { max_tokens: 256 } };
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -381,7 +381,7 @@ describe("applyExtraParamsToAgent", () => {
    const payloads: Record<string, unknown>[] = [];
    const baseStreamFn: StreamFn = (_model, _context, options) => {
      const payload: Record<string, unknown> = { reasoning_effort: "medium" };
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -588,7 +588,7 @@ describe("applyExtraParamsToAgent", () => {
    const payloads: Record<string, unknown>[] = [];
    const baseStreamFn: StreamFn = (_model, _context, options) => {
      const payload: Record<string, unknown> = { thinking: "off" };
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -619,7 +619,7 @@ describe("applyExtraParamsToAgent", () => {
    const payloads: Record<string, unknown>[] = [];
    const baseStreamFn: StreamFn = (_model, _context, options) => {
      const payload: Record<string, unknown> = { thinking: "off" };
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -650,7 +650,7 @@ describe("applyExtraParamsToAgent", () => {
    const payloads: Record<string, unknown>[] = [];
    const baseStreamFn: StreamFn = (_model, _context, options) => {
      const payload: Record<string, unknown> = {};
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -674,7 +674,7 @@ describe("applyExtraParamsToAgent", () => {
    const payloads: Record<string, unknown>[] = [];
    const baseStreamFn: StreamFn = (_model, _context, options) => {
      const payload: Record<string, unknown> = { tool_choice: "required" };
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -699,7 +699,7 @@ describe("applyExtraParamsToAgent", () => {
    const payloads: Record<string, unknown>[] = [];
    const baseStreamFn: StreamFn = (_model, _context, options) => {
      const payload: Record<string, unknown> = {};
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -749,7 +749,7 @@ describe("applyExtraParamsToAgent", () => {
        ],
        tool_choice: { type: "tool", name: "read" },
      };
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -793,7 +793,7 @@ describe("applyExtraParamsToAgent", () => {
          },
        ],
      };
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -832,7 +832,7 @@ describe("applyExtraParamsToAgent", () => {
          },
        ],
      };
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -896,7 +896,7 @@ describe("applyExtraParamsToAgent", () => {
          },
        },
      };
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
@ -943,7 +943,7 @@ describe("applyExtraParamsToAgent", () => {
          },
        },
      };
-      options?.onPayload?.(payload, model);
+      options?.onPayload?.(payload);
      payloads.push(payload);
      return {} as ReturnType<StreamFn>;
    };
--- a/src/agents/pi-embedded-runner/anthropic-stream-wrappers.ts
+++ b/src/agents/pi-embedded-runner/anthropic-stream-wrappers.ts
@ -298,7 +298,7 @@ export function createAnthropicToolPayloadCompatibilityWrapper(
            );
          }
        }
-        return originalOnPayload?.(payload, model);
+        return originalOnPayload?.(payload);
      },
    });
  };
--- a/src/agents/pi-embedded-runner/compact.ts
+++ b/src/agents/pi-embedded-runner/compact.ts
@ -114,6 +114,8 @@ export type CompactEmbeddedPiSessionParams = {
  /** Whether the sender is an owner (required for owner-only tools). */
  senderIsOwner?: boolean;
  sessionFile: string;
+  /** Optional caller-observed live prompt tokens used for compaction diagnostics. */
+  currentTokenCount?: number;
  workspaceDir: string;
  agentDir?: string;
  config?: OpenClawConfig;
@ -152,6 +154,12 @@ function createCompactionDiagId(): string {
  return `cmp-${Date.now().toString(36)}-${generateSecureToken(4)}`;
 }

+function normalizeObservedTokenCount(value: unknown): number | undefined {
+  return typeof value === "number" && Number.isFinite(value) && value > 0
+    ? Math.floor(value)
+    : undefined;
+}
+
 function getMessageTextChars(msg: AgentMessage): number {
  const content = (msg as { content?: unknown }).content;
  if (typeof content === "string") {
@ -228,6 +236,9 @@ function classifyCompactionReason(reason?: string): string {
  if (text.includes("already compacted")) {
    return "already_compacted_recently";
  }
+  if (text.includes("still exceeds target")) {
+    return "live_context_still_exceeds_target";
+  }
  if (text.includes("guard")) {
    return "guard_blocked";
  }
@ -701,6 +712,7 @@ export async function compactEmbeddedPiSessionDirect(
        const missingSessionKey = !params.sessionKey || !params.sessionKey.trim();
        const hookSessionKey = params.sessionKey?.trim() || params.sessionId;
        const hookRunner = getGlobalHookRunner();
+        const observedTokenCount = normalizeObservedTokenCount(params.currentTokenCount);
        const messageCountOriginal = originalMessages.length;
        let tokenCountOriginal: number | undefined;
        try {
@ -712,14 +724,16 @@ export async function compactEmbeddedPiSessionDirect(
          tokenCountOriginal = undefined;
        }
        const messageCountBefore = session.messages.length;
-        let tokenCountBefore: number | undefined;
-        try {
-          tokenCountBefore = 0;
-          for (const message of session.messages) {
-            tokenCountBefore += estimateTokens(message);
+        let tokenCountBefore = observedTokenCount;
+        if (tokenCountBefore === undefined) {
+          try {
+            tokenCountBefore = 0;
+            for (const message of session.messages) {
+              tokenCountBefore += estimateTokens(message);
+            }
+          } catch {
+            tokenCountBefore = undefined;
          }
-        } catch {
-          tokenCountBefore = undefined;
        }
        // TODO(#7175): Consider exposing full message snapshots or pre-compaction injection
        // hooks; current events only report counts/metadata.
@ -802,7 +816,7 @@ export async function compactEmbeddedPiSessionDirect(
            tokensAfter += estimateTokens(message);
          }
          // Sanity check: tokensAfter should be less than tokensBefore
-          if (tokensAfter > result.tokensBefore) {
+          if (tokensAfter > (observedTokenCount ?? result.tokensBefore)) {
            tokensAfter = undefined; // Don't trust the estimate
          }
        } catch {
@ -876,7 +890,7 @@ export async function compactEmbeddedPiSessionDirect(
          result: {
            summary: result.summary,
            firstKeptEntryId: result.firstKeptEntryId,
-            tokensBefore: result.tokensBefore,
+            tokensBefore: observedTokenCount ?? result.tokensBefore,
            tokensAfter,
            details: result.details,
          },
@ -977,6 +991,7 @@ export async function compactEmbeddedPiSession(
          sessionId: params.sessionId,
          sessionFile: params.sessionFile,
          tokenBudget: ceCtxInfo.tokens,
+          currentTokenCount: params.currentTokenCount,
          customInstructions: params.customInstructions,
          force: params.trigger === "manual",
          runtimeContext: params as Record<string, unknown>,
--- a/src/agents/pi-embedded-runner/extra-params.kilocode.test.ts
+++ b/src/agents/pi-embedded-runner/extra-params.kilocode.test.ts
@ -19,7 +19,7 @@ function applyAndCapture(params: {

  const baseStreamFn: StreamFn = (_model, _context, options) => {
    captured.headers = options?.headers;
-    options?.onPayload?.({}, _model);
+    options?.onPayload?.({});
    return createAssistantMessageEventStream();
  };
  const agent = { streamFn: baseStreamFn };
@ -97,7 +97,7 @@ describe("extra-params: Kilocode kilo/auto reasoning", () => {

    const baseStreamFn: StreamFn = (_model, _context, options) => {
      const payload: Record<string, unknown> = { reasoning_effort: "high" };
-      options?.onPayload?.(payload, _model);
+      options?.onPayload?.(payload);
      capturedPayload = payload;
      return createAssistantMessageEventStream();
    };
@ -125,7 +125,7 @@ describe("extra-params: Kilocode kilo/auto reasoning", () => {

    const baseStreamFn: StreamFn = (_model, _context, options) => {
      const payload: Record<string, unknown> = {};
-      options?.onPayload?.(payload, _model);
+      options?.onPayload?.(payload);
      capturedPayload = payload;
      return createAssistantMessageEventStream();
    };
@ -158,7 +158,7 @@ describe("extra-params: Kilocode kilo/auto reasoning", () => {

    const baseStreamFn: StreamFn = (_model, _context, options) => {
      const payload: Record<string, unknown> = { reasoning_effort: "high" };
-      options?.onPayload?.(payload, _model);
+      options?.onPayload?.(payload);
      capturedPayload = payload;
      return createAssistantMessageEventStream();
    };
--- a/src/agents/pi-embedded-runner/extra-params.openrouter-cache-control.test.ts
+++ b/src/agents/pi-embedded-runner/extra-params.openrouter-cache-control.test.ts
@ -13,7 +13,7 @@ type StreamPayload = {

 function runOpenRouterPayload(payload: StreamPayload, modelId: string) {
  const baseStreamFn: StreamFn = (_model, _context, options) => {
-    options?.onPayload?.(payload, _model);
+    options?.onPayload?.(payload);
    return createAssistantMessageEventStream();
  };
  const agent = { streamFn: baseStreamFn };
--- a/src/agents/pi-embedded-runner/extra-params.ts
+++ b/src/agents/pi-embedded-runner/extra-params.ts
@ -230,7 +230,7 @@ function createGoogleThinkingPayloadWrapper(
            thinkingLevel,
          });
        }
-        return onPayload?.(payload, model);
+        return onPayload?.(payload);
      },
    });
  };
@ -263,7 +263,7 @@ function createZaiToolStreamWrapper(
          // Inject tool_stream: true for Z.AI API
          (payload as Record<string, unknown>).tool_stream = true;
        }
-        return originalOnPayload?.(payload, model);
+        return originalOnPayload?.(payload);
      },
    });
  };
@ -310,7 +310,7 @@ function createParallelToolCallsWrapper(
        if (payload && typeof payload === "object") {
          (payload as Record<string, unknown>).parallel_tool_calls = enabled;
        }
-        return originalOnPayload?.(payload, model);
+        return originalOnPayload?.(payload);
      },
    });
  };
--- a/src/agents/pi-embedded-runner/extra-params.zai-tool-stream.test.ts
+++ b/src/agents/pi-embedded-runner/extra-params.zai-tool-stream.test.ts
@ -22,7 +22,7 @@ type ToolStreamCase = {
 function runToolStreamCase(params: ToolStreamCase) {
  const payload: Record<string, unknown> = { model: params.model.id, messages: [] };
  const baseStreamFn: StreamFn = (model, _context, options) => {
-    options?.onPayload?.(payload, model);
+    options?.onPayload?.(payload);
    return {} as ReturnType<StreamFn>;
  };
  const agent = { streamFn: baseStreamFn };
--- a/src/agents/pi-embedded-runner/moonshot-stream-wrappers.ts
+++ b/src/agents/pi-embedded-runner/moonshot-stream-wrappers.ts
@ -60,7 +60,7 @@ export function createSiliconFlowThinkingWrapper(baseStreamFn: StreamFn | undefi
            payloadObj.thinking = null;
          }
        }
-        return originalOnPayload?.(payload, model);
+        return originalOnPayload?.(payload);
      },
    });
  };
@ -106,7 +106,7 @@ export function createMoonshotThinkingWrapper(
            payloadObj.tool_choice = "auto";
          }
        }
-        return originalOnPayload?.(payload, model);
+        return originalOnPayload?.(payload);
      },
    });
  };
--- a/src/agents/pi-embedded-runner/openai-stream-wrappers.ts
+++ b/src/agents/pi-embedded-runner/openai-stream-wrappers.ts
@ -197,7 +197,7 @@ export function createOpenAIResponsesContextManagementWrapper(
            compactThreshold,
          });
        }
-        return originalOnPayload?.(payload, model);
+        return originalOnPayload?.(payload);
      },
    });
  };
@ -226,7 +226,7 @@ export function createOpenAIServiceTierWrapper(
            payloadObj.service_tier = serviceTier;
          }
        }
-        return originalOnPayload?.(payload, model);
+        return originalOnPayload?.(payload);
      },
    });
  };
--- a/src/agents/pi-embedded-runner/proxy-stream-wrappers.ts
+++ b/src/agents/pi-embedded-runner/proxy-stream-wrappers.ts
@ -92,7 +92,7 @@ export function createOpenRouterSystemCacheWrapper(baseStreamFn: StreamFn | unde
            }
          }
        }
-        return originalOnPayload?.(payload, model);
+        return originalOnPayload?.(payload);
      },
    });
  };
@ -113,7 +113,7 @@ export function createOpenRouterWrapper(
      },
      onPayload: (payload) => {
        normalizeProxyReasoningPayload(payload, thinkingLevel);
-        return onPayload?.(payload, model);
+        return onPayload?.(payload);
      },
    });
  };
@ -138,7 +138,7 @@ export function createKilocodeWrapper(
      },
      onPayload: (payload) => {
        normalizeProxyReasoningPayload(payload, thinkingLevel);
-        return onPayload?.(payload, model);
+        return onPayload?.(payload);
      },
    });
  };
--- a/src/agents/pi-embedded-runner/run.overflow-compaction.mocks.shared.ts
+++ b/src/agents/pi-embedded-runner/run.overflow-compaction.mocks.shared.ts
@ -109,13 +109,21 @@ vi.mock("../workspace-run.js", () => ({
 vi.mock("../pi-embedded-helpers.js", () => ({
  formatBillingErrorMessage: vi.fn(() => ""),
  classifyFailoverReason: vi.fn(() => null),
+  extractObservedOverflowTokenCount: vi.fn((msg?: string) => {
+    const match = msg?.match(/prompt is too long:\s*([\d,]+)\s+tokens\s*>\s*[\d,]+\s+maximum/i);
+    return match?.[1] ? Number(match[1].replaceAll(",", "")) : undefined;
+  }),
  formatAssistantErrorText: vi.fn(() => ""),
  isAuthAssistantError: vi.fn(() => false),
  isBillingAssistantError: vi.fn(() => false),
  isCompactionFailureError: vi.fn(() => false),
  isLikelyContextOverflowError: vi.fn((msg?: string) => {
    const lower = (msg ?? "").toLowerCase();
-    return lower.includes("request_too_large") || lower.includes("context window exceeded");
+    return (
+      lower.includes("request_too_large") ||
+      lower.includes("context window exceeded") ||
+      lower.includes("prompt is too long")
+    );
  }),
  isFailoverAssistantError: vi.fn(() => false),
  isFailoverErrorMessage: vi.fn(() => false),
--- a/src/agents/pi-embedded-runner/run.overflow-compaction.test.ts
+++ b/src/agents/pi-embedded-runner/run.overflow-compaction.test.ts
@ -111,6 +111,32 @@ describe("runEmbeddedPiAgent overflow compaction trigger routing", () => {
    );
  });

+  it("passes observed overflow token counts into compaction when providers report them", async () => {
+    const overflowError = new Error(
+      '400 {"type":"error","error":{"type":"invalid_request_error","message":"prompt is too long: 277403 tokens > 200000 maximum"}}',
+    );
+
+    mockedRunEmbeddedAttempt
+      .mockResolvedValueOnce(makeAttemptResult({ promptError: overflowError }))
+      .mockResolvedValueOnce(makeAttemptResult({ promptError: null }));
+    mockedCompactDirect.mockResolvedValueOnce(
+      makeCompactionSuccess({
+        summary: "Compacted session",
+        firstKeptEntryId: "entry-8",
+        tokensBefore: 277403,
+      }),
+    );
+
+    const result = await runEmbeddedPiAgent(overflowBaseRunParams);
+
+    expect(mockedCompactDirect).toHaveBeenCalledWith(
+      expect.objectContaining({
+        currentTokenCount: 277403,
+      }),
+    );
+    expect(result.meta.error).toBeUndefined();
+  });
+
  it("does not reset compaction attempt budget after successful tool-result truncation", async () => {
    const overflowError = queueOverflowAttemptWithOversizedToolOutput(
      mockedRunEmbeddedAttempt,
--- a/src/agents/pi-embedded-runner/run.ts
+++ b/src/agents/pi-embedded-runner/run.ts
@ -40,6 +40,7 @@ import { ensureOpenClawModelsJson } from "../models-config.js";
 import {
  formatBillingErrorMessage,
  classifyFailoverReason,
+  extractObservedOverflowTokenCount,
  formatAssistantErrorText,
  isAuthAssistantError,
  isBillingAssistantError,
@ -988,11 +989,13 @@ export async function runEmbeddedPiAgent(
            const overflowDiagId = createCompactionDiagId();
            const errorText = contextOverflowError.text;
            const msgCount = attempt.messagesSnapshot?.length ?? 0;
+            const observedOverflowTokens = extractObservedOverflowTokenCount(errorText);
            log.warn(
              `[context-overflow-diag] sessionKey=${params.sessionKey ?? params.sessionId} ` +
                `provider=${provider}/${modelId} source=${contextOverflowError.source} ` +
                `messages=${msgCount} sessionFile=${params.sessionFile} ` +
                `diagId=${overflowDiagId} compactionAttempts=${overflowCompactionAttempts} ` +
+                `observedTokens=${observedOverflowTokens ?? "unknown"} ` +
                `error=${errorText.slice(0, 200)}`,
            );
            const isCompactionFailure = isCompactionFailureError(errorText);
@ -1052,6 +1055,9 @@ export async function runEmbeddedPiAgent(
                  sessionId: params.sessionId,
                  sessionFile: params.sessionFile,
                  tokenBudget: ctxInfo.tokens,
+                  ...(observedOverflowTokens !== undefined
+                    ? { currentTokenCount: observedOverflowTokens }
+                    : {}),
                  force: true,
                  compactionTarget: "budget",
                  runtimeContext: {
@ -1074,6 +1080,9 @@ export async function runEmbeddedPiAgent(
                    extraSystemPrompt: params.extraSystemPrompt,
                    ownerNumbers: params.ownerNumbers,
                    trigger: "overflow",
+                    ...(observedOverflowTokens !== undefined
+                      ? { currentTokenCount: observedOverflowTokens }
+                      : {}),
                    diagId: overflowDiagId,
                    attempt: overflowCompactionAttempts,
                    maxAttempts: MAX_OVERFLOW_COMPACTION_ATTEMPTS,
--- a/src/agents/pi-embedded-runner/run/attempt.ts
+++ b/src/agents/pi-embedded-runner/run/attempt.ts
@ -233,14 +233,14 @@ export function wrapOllamaCompatNumCtx(baseFn: StreamFn | undefined, numCtx: num
      ...options,
      onPayload: (payload: unknown) => {
        if (!payload || typeof payload !== "object") {
-          return options?.onPayload?.(payload, model);
+          return options?.onPayload?.(payload);
        }
        const payloadRecord = payload as Record<string, unknown>;
        if (!payloadRecord.options || typeof payloadRecord.options !== "object") {
          payloadRecord.options = {};
        }
        (payloadRecord.options as Record<string, unknown>).num_ctx = numCtx;
-        return options?.onPayload?.(payload, model);
+        return options?.onPayload?.(payload);
      },
    });
 }
--- a/src/commands/openai-codex-oauth.test.ts
+++ b/src/commands/openai-codex-oauth.test.ts
@ -9,7 +9,7 @@ const mocks = vi.hoisted(() => ({
  formatOpenAIOAuthTlsPreflightFix: vi.fn(),
 }));

-vi.mock("@mariozechner/pi-ai/oauth", () => ({
+vi.mock("@mariozechner/pi-ai", () => ({
  loginOpenAICodex: mocks.loginOpenAICodex,
 }));

--- a/src/commands/openai-codex-oauth.ts
+++ b/src/commands/openai-codex-oauth.ts
@ -1,5 +1,5 @@
 import type { OAuthCredentials } from "@mariozechner/pi-ai";
-import { loginOpenAICodex } from "@mariozechner/pi-ai/oauth";
+import { loginOpenAICodex } from "@mariozechner/pi-ai";
 import type { RuntimeEnv } from "../runtime.js";
 import type { WizardPrompter } from "../wizard/prompts.js";
 import { createVpsAwareOAuthHandlers } from "./oauth-flow.js";
--- a/src/context-engine/context-engine.test.ts
+++ b/src/context-engine/context-engine.test.ts
@ -1,5 +1,6 @@
 import type { AgentMessage } from "@mariozechner/pi-agent-core";
-import { describe, expect, it, beforeEach } from "vitest";
+import { beforeEach, describe, expect, it, vi } from "vitest";
+import { compactEmbeddedPiSessionDirect } from "../agents/pi-embedded-runner/compact.runtime.js";
 // ---------------------------------------------------------------------------
 // We dynamically import the registry so we can get a fresh module per test
 // group when needed.  For most groups we use the shared singleton directly.
@ -19,6 +20,23 @@ import type {
  IngestResult,
 } from "./types.js";

+vi.mock("../agents/pi-embedded-runner/compact.runtime.js", () => ({
+  compactEmbeddedPiSessionDirect: vi.fn(async () => ({
+    ok: true,
+    compacted: false,
+    reason: "mock compaction",
+    result: {
+      summary: "",
+      firstKeptEntryId: "",
+      tokensBefore: 0,
+      tokensAfter: 0,
+      details: undefined,
+    },
+  })),
+}));
+
+const mockedCompactEmbeddedPiSessionDirect = vi.mocked(compactEmbeddedPiSessionDirect);
+
 // ---------------------------------------------------------------------------
 // Helpers
 // ---------------------------------------------------------------------------
@ -91,6 +109,10 @@ class MockContextEngine implements ContextEngine {
 // ═══════════════════════════════════════════════════════════════════════════

 describe("Engine contract tests", () => {
+  beforeEach(() => {
+    mockedCompactEmbeddedPiSessionDirect.mockClear();
+  });
+
  it("a mock engine implementing ContextEngine can be registered and resolved", async () => {
    const factory = () => new MockContextEngine();
    registerContextEngine("mock", factory);
@ -153,6 +175,25 @@ describe("Engine contract tests", () => {
    // Should complete without error
    await expect(engine.dispose()).resolves.toBeUndefined();
  });
+
+  it("legacy compact preserves runtimeContext currentTokenCount when top-level value is absent", async () => {
+    const engine = new LegacyContextEngine();
+
+    await engine.compact({
+      sessionId: "s1",
+      sessionFile: "/tmp/session.json",
+      runtimeContext: {
+        workspaceDir: "/tmp/workspace",
+        currentTokenCount: 277403,
+      },
+    });
+
+    expect(mockedCompactEmbeddedPiSessionDirect).toHaveBeenCalledWith(
+      expect.objectContaining({
+        currentTokenCount: 277403,
+      }),
+    );
+  });
 });

 // ═══════════════════════════════════════════════════════════════════════════
--- a/src/context-engine/legacy.ts
+++ b/src/context-engine/legacy.ts
@ -78,6 +78,13 @@ export class LegacyContextEngine implements ContextEngine {
    // set by the caller in run.ts. We spread them and override the fields
    // that come from the ContextEngine compact() signature directly.
    const runtimeContext = params.runtimeContext ?? {};
+    const currentTokenCount =
+      params.currentTokenCount ??
+      (typeof runtimeContext.currentTokenCount === "number" &&
+      Number.isFinite(runtimeContext.currentTokenCount) &&
+      runtimeContext.currentTokenCount > 0
+        ? Math.floor(runtimeContext.currentTokenCount)
+        : undefined);

    // eslint-disable-next-line @typescript-eslint/no-explicit-any -- bridge runtimeContext matches CompactEmbeddedPiSessionParams
    const result = await compactEmbeddedPiSessionDirect({
@ -85,6 +92,7 @@ export class LegacyContextEngine implements ContextEngine {
      sessionId: params.sessionId,
      sessionFile: params.sessionFile,
      tokenBudget: params.tokenBudget,
+      ...(currentTokenCount !== undefined ? { currentTokenCount } : {}),
      force: params.force,
      customInstructions: params.customInstructions,
      workspaceDir: (runtimeContext.workspaceDir as string) ?? process.cwd(),
--- a/test/setup.ts
+++ b/test/setup.ts
@ -10,12 +10,6 @@ vi.mock("@mariozechner/pi-ai", async (importOriginal) => {
  };
 });

-vi.mock("@mariozechner/pi-ai/oauth", () => ({
-  getOAuthApiKey: () => undefined,
-  getOAuthProviders: () => [],
-  loginOpenAICodex: vi.fn(),
-}));
-
 // Ensure Vitest environment is properly set
 process.env.VITEST = "true";
 // Config validation walks plugin manifests; keep an aggressive cache in tests to avoid