fix: improve context overflow error with diagnostic details by superlowburn · Pull Request #15749 · openclaw/openclaw

superlowburn · 2026-02-13T20:36:25Z

Summary

Replaces the generic "Context overflow: prompt too large" message with diagnostic details
Now includes provider/model name, context window limit, session identifier, and message count
Adds actionable recovery steps (run /reset, switch model, adjust contextTokens)

Before:

Context overflow: prompt too large for the model. Try /reset (or /new) to start a fresh session, or use a larger-context model.

After:

Context overflow (anthropic/claude-opus-4-5, limit: 200,000 tokens)
  Session: discord:channel:123456789
  Messages: 47

Actions:
  • Run /reset to clear context and start fresh
  • Or switch to a model with a larger context window
  • Consider lowering contextTokens in config to trigger earlier compaction

Closes #9409

Test plan

Lint passes (oxlint, no warnings/errors)
Existing e2e tests check isError: true and error.kind, not exact text — unchanged
Manual test: trigger a context overflow and verify the new message format

🤖 Generated with Claude Code

Greptile Overview

Greptile Summary

This change updates the embedded PI agent’s context-overflow handling to return a multi-line error message with diagnostics (provider/model, context-window limit, session label, message count) and suggested recovery actions instead of the previous generic one-liner.

The behavior lives in src/agents/pi-embedded-runner/run.ts inside the overflow/compaction recovery branch, and affects what end users see when the runner detects a context overflow and can’t recover via auto-compaction/tool-result truncation.

Confidence Score: 3/5

This PR is close to safe to merge but has a couple of user-facing error-handling edge cases to fix first.
Core change is localized and straightforward, but the new formatting can throw if the context limit isn’t a number, and it may expose sensitive session identifiers and mislabel compaction failures as context overflows.
src/agents/pi-embedded-runner/run.ts

_{Last reviewed commit: 281f021}

Include provider/model, context window limit, session identifier, and message count in the overflow error. Add actionable recovery steps (/reset, switch model, adjust contextTokens). Closes openclaw#9409 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

greptile-apps

_{1 file reviewed, 3 comments}

_{Edit Code Review Agent Settings | Greptile}

greptile-apps · 2026-02-13T20:40:54Z

src/agents/pi-embedded-runner/run.ts

            const kind = isCompactionFailure ? "compaction_failure" : "context_overflow";
+            const contextLimit = ctxInfo.tokens;
+            const sessionLabel = params.sessionKey ?? params.sessionId;
+            const overflowDetails = [


Possible crash on overflow

ctxInfo.tokens is used as a number (contextLimit.toLocaleString()), but resolveContextWindowInfo isn’t guaranteed to return a numeric tokens value in all configurations. If tokens is undefined/non-number, this will throw while handling a context overflow, masking the real error and preventing an error payload from being returned.

Consider guarding/normalizing ctxInfo.tokens before calling toLocaleString().

Prompt To Fix With AI

This is a comment left during a code review. Path: src/agents/pi-embedded-runner/run.ts Line: 603:606 Comment: **Possible crash on overflow** `ctxInfo.tokens` is used as a number (`contextLimit.toLocaleString()`), but `resolveContextWindowInfo` isn’t guaranteed to return a numeric `tokens` value in all configurations. If `tokens` is `undefined`/non-number, this will throw while handling a context overflow, masking the real error and preventing an error payload from being returned. Consider guarding/normalizing `ctxInfo.tokens` before calling `toLocaleString()`. How can I resolve this? If you propose a fix, please make it concise.

Valid — ctxInfo.tokens could be undefined depending on the provider/config, which would throw on .toLocaleString(). Will add a fallback: const contextLimit = ctxInfo.tokens ?? 'unknown' and skip the .toLocaleString() when it's not a number.

greptile-apps · 2026-02-13T20:40:55Z

src/agents/pi-embedded-runner/run.ts

+            const sessionLabel = params.sessionKey ?? params.sessionId;
+            const overflowDetails = [
+              `Context overflow (${provider}/${modelId}, limit: ${contextLimit.toLocaleString()} tokens)`,
+              sessionLabel ? `  Session: ${sessionLabel}` : null,
+              `  Messages: ${msgCount}`,


Leaks raw session identifier

The user-facing overflow text now includes Session: ${params.sessionKey ?? params.sessionId}. Elsewhere in this file session identifiers are redacted before logging (redactRunIdentifier(...)), so this appears to be sensitive/internal data that shouldn’t be echoed back to end users (especially if sessionId encodes channel IDs, phone numbers, etc.).

If the goal is diagnostics, consider either redacting here too or only including a short/hashed label intended for users.

Prompt To Fix With AI

This is a comment left during a code review. Path: src/agents/pi-embedded-runner/run.ts Line: 605:609 Comment: **Leaks raw session identifier** The user-facing overflow text now includes `Session: ${params.sessionKey ?? params.sessionId}`. Elsewhere in this file session identifiers are redacted before logging (`redactRunIdentifier(...)`), so this appears to be sensitive/internal data that shouldn’t be echoed back to end users (especially if `sessionId` encodes channel IDs, phone numbers, etc.). If the goal is diagnostics, consider either redacting here too or only including a short/hashed label intended for users. How can I resolve this? If you propose a fix, please make it concise.

Good catch — the rest of this file redacts session identifiers before logging. Will remove the session line from the user-facing error. Diagnostic details should go to logs, not to the user.

greptile-apps · 2026-02-13T20:40:56Z

src/agents/pi-embedded-runner/run.ts

+            const overflowDetails = [
+              `Context overflow (${provider}/${modelId}, limit: ${contextLimit.toLocaleString()} tokens)`,
+              sessionLabel ? `  Session: ${sessionLabel}` : null,
+              `  Messages: ${msgCount}`,
+              "",
+              "Actions:",
+              "  \u2022 Run /reset to clear context and start fresh",
+              "  \u2022 Or switch to a model with a larger context window",
+              "  \u2022 Consider lowering contextTokens in config to trigger earlier compaction",


Compaction failures misreported

This new message is returned for both context_overflow and compaction_failure (via kind), but the header always says Context overflow (...) and the suggested actions are overflow-specific. When isCompactionFailure is true (e.g., summarization/compaction request itself overflows), this text is misleading.

Consider branching the header/actions for compaction_failure vs context_overflow so users get the right recovery steps.

Prompt To Fix With AI

This is a comment left during a code review. Path: src/agents/pi-embedded-runner/run.ts Line: 606:614 Comment: **Compaction failures misreported** This new message is returned for both `context_overflow` and `compaction_failure` (via `kind`), but the header always says `Context overflow (...)` and the suggested actions are overflow-specific. When `isCompactionFailure` is true (e.g., summarization/compaction request itself overflows), this text is misleading. Consider branching the header/actions for `compaction_failure` vs `context_overflow` so users get the right recovery steps. How can I resolve this? If you propose a fix, please make it concise.

Valid — the kind variable distinguishes the two cases but the user-facing message doesn't. Will add a conditional header: "Context overflow" vs "Compaction failure" based on isCompactionFailure, with appropriate actions for each.

…ion vs overflow Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…paction vs overflow header Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

openclaw-barnacle · 2026-02-21T04:00:21Z

This pull request has been automatically marked as stale due to inactivity.
Please add updates or it will be closed.

openclaw-barnacle bot added agents Agent runtime and tooling size: XS labels Feb 13, 2026

greptile-apps bot reviewed Feb 13, 2026

View reviewed changes

superlowburn and others added 2 commits February 13, 2026 22:40

fix: defensive tokens check, redact session ID, differentiate compact…

da24eb5

…ion vs overflow Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

fix: guard undefined tokens, remove session ID from error, branch com…

d458d20

…paction vs overflow header Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

thewilloftheshadow force-pushed the main branch from bfc1ccb to f92900f Compare February 15, 2026 18:46

This was referenced Feb 17, 2026

[Bug]: Context-overflow recovery can finalize empty instead of deterministic fallback #19550

Closed

fix(reply): make overflow fallback deterministic for empty recovery payloads #19551

Draft

openclaw-barnacle bot added stale Marked as stale due to inactivity and removed stale Marked as stale due to inactivity labels Feb 21, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix: improve context overflow error with diagnostic details#15749

fix: improve context overflow error with diagnostic details#15749
superlowburn wants to merge 3 commits intoopenclaw:mainfrom
superlowburn:fix/9409-context-overflow-error-message

superlowburn commented Feb 13, 2026 •

edited by greptile-apps bot

Loading

Uh oh!

greptile-apps bot left a comment

Uh oh!

greptile-apps bot Feb 13, 2026

Uh oh!

superlowburn Feb 14, 2026

Uh oh!

greptile-apps bot Feb 13, 2026

Uh oh!

superlowburn Feb 14, 2026

Uh oh!

greptile-apps bot Feb 13, 2026

Uh oh!

superlowburn Feb 14, 2026

Uh oh!

openclaw-barnacle bot commented Feb 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

superlowburn commented Feb 13, 2026 • edited by greptile-apps bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Greptile Overview

Greptile Summary

Confidence Score: 3/5

Uh oh!

greptile-apps bot left a comment

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Feb 13, 2026

Choose a reason for hiding this comment

Uh oh!

superlowburn Feb 14, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Feb 13, 2026

Choose a reason for hiding this comment

Uh oh!

superlowburn Feb 14, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Feb 13, 2026

Choose a reason for hiding this comment

Uh oh!

superlowburn Feb 14, 2026

Choose a reason for hiding this comment

Uh oh!

openclaw-barnacle bot commented Feb 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

superlowburn commented Feb 13, 2026 •

edited by greptile-apps bot

Loading