fix: implement boot-time patch for Anthropic thinking blocks API signature (#27825) by ishan8351 · Pull Request #27965 · openclaw/openclaw

ishan8351 · 2026-02-26T22:23:52Z

Summary

Describe the problem and fix in 2–5 bullets:

Problem: The sanitizeSurrogates() utility in @mariozechner/pi-ai mutates Anthropic thinking blocks, which breaks cryptographic signatures during conversation history replay and causes session stalls.
Why it matters: Users with long-running sessions (e.g., heartbeat agents) experience permanent data corruption when compacted context is saved with invalid signatures.
What changed: Added a boot-time runtime patch in openclaw.mjs. This dynamically removes the sanitizer wrapper using regex before the app boots.
What did NOT change (scope boundary): Sanitization remains active for standard text blocks and system prompts where signature integrity isn't a factor.

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

Closes sanitizeSurrogates() corrupts thinking block content before API replay #27825

User-visible / Behavior Changes

List user-visible changes (including defaults/config).
Displays patch applied message on boot when debug mode (OPENCLAW_DEBUG=1) is enabled, and prints a helpful warning if file permissions block the patch.

Security Impact (required)

New permissions/capabilities? No
Secrets/tokens handling changed? No
New/changed network calls? No
Command/tool execution surface changed? No
Data access scope changed? No

Repro + Verification

Environment

OS: WSL
Runtime/container: pnpm (Node.js)
Model/provider: Anthropic
Integration/channel (if any): Telegram/OpenClaw session replay

Steps

1.Enable "Extended Thinking" for a Claude-based assistant.
2.Generate a response that includes a reasoning trace (thinking block).
3.Send a follow-up message to trigger a history replay.

Expected

-The session continues normally as the thinking block is sent back to the API unmodified.

Actual

-The API returns an error or stalls because the thinking block content was mutated by the surrogate sanitizer, breaking the signature.

Human Verification (required)

What you personally verified (not just CI), and how:

Verified scenarios: Manually verified that the patch runs and has desired effects.
Edge cases checked: Verified that standard text blocks still undergo sanitization.
What you did not verify: Did not verify performance overhead (should be negligible).

Compatibility / Migration

Backward compatible? Yes
Config/env changes? No
Migration needed? No
If yes, exact upgrade steps: None. Automatic execution on initiation and works universally

Failure Recovery (if this breaks)

How to disable/revert this change quickly: Remove the applyAnthropicThinkingPatch() block from the top of openclaw.mjs.
Known bad symptoms reviewers should watch for: If the upstream pi-ai library drastically rewrites its file structure or code in a future update, the Regex won't match. The patch will safely skip itself without crashing, but users will revert to the "stalled session" bug. Reviewers can watch the boot logs for [OpenClaw Patch] Skipped patch: to diagnose.

Risks and Mitigations

List only real risks for this PR. Add/remove entries as needed. If none, write None.

Risk: Deployments in strictly read-only environments (e.g., some Docker containers or serverless functions) will block the runtime patch from modifying node_modules.
-Mitigation: Caught EROFS and EACCES errors in the boot script. It will gracefully skip the patch and log a clear warning rather than crashing the app, leaving the user no worse off than before the PR

greptile-apps · 2026-02-26T22:25:39Z

Greptile Summary

This PR applies a pnpm patch to bypass sanitizeSurrogates() for Anthropic thinking blocks during conversation history replay. The surrogate sanitizer was modifying thinking block content, breaking the API signature and causing session stalls. The fix removes sanitization only for thinking blocks (both with and without signatures) while preserving it for standard text blocks.

Key changes:

Added patchedDependencies entry in package.json for @mariozechner/pi-ai
Created patch file that removes two sanitizeSurrogates() calls in dist/providers/anthropic.js
Updated pnpm-lock.yaml to reference the patch with its hash

The approach is minimal and targeted - only the specific lines causing the issue are modified, and the patch correctly preserves API signature integrity for Anthropic's extended thinking feature.

Confidence Score: 4/5

Safe to merge with minor verification needed
The patch is minimal, well-scoped, and addresses a specific bug. The dependency version is correctly specified as an exact version per repository guidelines. The only considerations are: (1) this modifies a third-party dependency's dist files which could be fragile across updates, and (2) the PR author has manually verified the fix but automated testing coverage for this specific scenario is unclear.
No files require special attention - the changes are straightforward dependency patching

_{Last reviewed commit: 4c40946}

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4c40946441

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

patches/@mariozechner__pi-ai.patch

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ae2bbead6a

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

patches/@mariozechner__pi-ai.patch

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 05717f9eb0

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

package.json

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c51587bf5b

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

package.json

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c5e6a961a0

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

openclaw.mjs

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 52d4d21f12

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

openclaw.mjs

…ature (openclaw#27825)

ishan8351 · 2026-03-01T11:39:15Z

Regarding upstream @mariozechner/pi-ai
Came across this code snippet:
const client = new Anthropic({
apiKey: null,
authToken: apiKey,
baseURL: model.baseUrl,
dangerouslyAllowBrowser: true,
defaultHeaders: mergeHeaders({
accept: "application/json",
"anthropic-dangerous-direct-browser-access": "true",
"anthropic-beta": claude-code-20250219,oauth-2025-04-20,${betaFeatures.join(",")},
"user-agent": claude-cli/${claudeCodeVersion},
"x-app": "cli",
}, model.headers, optionsHeaders),
});

Looks like headers are hardcoded to spoof claude-cli. This violates Anthropic's Terms of Service, and could result in permanent OAuth/API bans for users.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3aea5df9be

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

openclaw.mjs

…27825) - Add sanitizePayload() with structural traversal - Bypass sanitization for signed thinking blocks - Preserve normal sanitization for other content - Config-driven toggle for flexible deployment - Replace runtime patching with clean code solution Supersedes openclaw#27965

openclaw-barnacle · 2026-03-31T04:25:51Z

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

openclaw-barnacle · 2026-04-04T04:09:04Z

Closing due to inactivity.
If you believe this PR should be revived, post in #pr-thunderdome-dangerzone on Discord to talk to a maintainer.
That channel is the escape hatch for high-quality PRs that get auto-closed.

openclaw-barnacle bot added the size: XS label Feb 26, 2026

chatgpt-codex-connector bot reviewed Feb 26, 2026

View reviewed changes

patches/@mariozechner__pi-ai.patch Outdated Show resolved Hide resolved

ishan8351 force-pushed the fix/issue-27825-skip-thinking-sanitization branch from 4c40946 to ae2bbea Compare February 27, 2026 04:22

chatgpt-codex-connector bot reviewed Feb 27, 2026

View reviewed changes

patches/@mariozechner__pi-ai.patch Outdated Show resolved Hide resolved

ishan8351 force-pushed the fix/issue-27825-skip-thinking-sanitization branch from ae2bbea to 05717f9 Compare February 27, 2026 04:29

chatgpt-codex-connector bot reviewed Feb 27, 2026

View reviewed changes

package.json Outdated Show resolved Hide resolved

ishan8351 force-pushed the fix/issue-27825-skip-thinking-sanitization branch from 05717f9 to c51587b Compare February 27, 2026 09:15

chatgpt-codex-connector bot reviewed Feb 27, 2026

View reviewed changes

package.json Outdated Show resolved Hide resolved

GodsBoy mentioned this pull request Feb 27, 2026

[Bug]: Anthropic API rejects "thinking blocks cannot be modified" in extended conversations #24612

Closed

hardyhwood mentioned this pull request Feb 27, 2026

fix(agents): drop thinking blocks for Anthropic API to prevent rejection #27931

Closed

5 tasks

ishan8351 force-pushed the fix/issue-27825-skip-thinking-sanitization branch from c51587b to c5e6a96 Compare March 1, 2026 10:55

ishan8351 changed the title ~~fix: skip anthropic thinking block sanitization to preserve API signature (#27825)~~ fix: implement boot-time patch for Anthropic thinking blocks API signature (#27825) Mar 1, 2026

chatgpt-codex-connector bot reviewed Mar 1, 2026

View reviewed changes

openclaw.mjs Outdated Show resolved Hide resolved

ishan8351 force-pushed the fix/issue-27825-skip-thinking-sanitization branch from c5e6a96 to 52d4d21 Compare March 1, 2026 11:11

openclaw-barnacle bot added size: S and removed size: XS labels Mar 1, 2026

chatgpt-codex-connector bot reviewed Mar 1, 2026

View reviewed changes

openclaw.mjs Show resolved Hide resolved

openclaw.mjs Outdated Show resolved Hide resolved

fix: implement boot-time patch for Anthropic thinking blocks API sign…

23fce86

…ature (openclaw#27825)

ishan8351 force-pushed the fix/issue-27825-skip-thinking-sanitization branch from 52d4d21 to 23fce86 Compare March 1, 2026 11:27

Merge branch 'main' into fix/issue-27825-skip-thinking-sanitization

3aea5df

chatgpt-codex-connector bot reviewed Mar 1, 2026

View reviewed changes

openclaw.mjs Show resolved Hide resolved

rikisann mentioned this pull request Mar 13, 2026

fix: boot-time patch for Anthropic thinking block signature corruption #44650

Open

Meli73 mentioned this pull request Mar 20, 2026

fix: patch @mariozechner/pi-ai to preserve signed thinking blocks (#27825) #51071

Closed

openclaw-barnacle bot added the stale Marked as stale due to inactivity label Mar 31, 2026

openclaw-barnacle bot closed this Apr 4, 2026

Uh oh!

Conversation

ishan8351 commented Feb 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

User-visible / Behavior Changes

Security Impact (required)

Repro + Verification

Environment

Steps

Expected

Actual

Human Verification (required)

Compatibility / Migration

Failure Recovery (if this breaks)

Risks and Mitigations

Uh oh!

greptile-apps bot commented Feb 26, 2026

Greptile Summary

Confidence Score: 4/5

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

ishan8351 commented Mar 1, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

openclaw-barnacle bot commented Mar 31, 2026

Uh oh!

openclaw-barnacle bot commented Apr 4, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

ishan8351 commented Feb 26, 2026 •

edited

Loading