Skip to content

fix(openclaw): strip <think/> tags from assistant messages#1280

Closed
hijzy wants to merge 1 commit intoMemTensor:mainfrom
hijzy:fix/1268-thinking-tags
Closed

fix(openclaw): strip <think/> tags from assistant messages#1280
hijzy wants to merge 1 commit intoMemTensor:mainfrom
hijzy:fix/1268-thinking-tags

Conversation

@hijzy
Copy link
Copy Markdown
Collaborator

@hijzy hijzy commented Mar 18, 2026

Problem

DeepSeek-style reasoning models emit <think…>…</think⟩ blocks when entering reasoning mode. The memos-local plugin does not recognize these tags, causing parse errors that trigger retry loops. This rapid retry cycle overloads OpenClaw and the AI model, eventually leading to timeouts.

Fix

Strip <think/> blocks early in the message capture pipeline (capture/index.ts), before any downstream parsing occurs. This is a minimal, non-breaking change — the thinking content is never stored or processed.

Closes #1268

DeepSeek-style reasoning models emit <think…>…</think⟩ blocks that
confuse downstream parsing and trigger error-retry loops, eventually
causing timeouts. Strip these tags early in the capture pipeline.

Closes MemTensor#1268
@hijzy hijzy marked this pull request as draft March 18, 2026 12:16
@hijzy hijzy closed this Mar 18, 2026
@hijzy hijzy deleted the fix/1268-thinking-tags branch March 24, 2026 02:21
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

fix: LLM思考标签混淆

1 participant