Claude Code CLI - API parsing and response issues

[bpeterson1991]

App Version

v3.23.18

API Provider

Anthropic

Model Used

Claude Code CLI - Sonnut and Opus

Roo Code Task Links (Optional)

https://app.roocode.com/share/ef6dbe36-04e6-4fba-8b71-425dd1536110
https://app.roocode.com/share/d04928d9-c217-4d81-a48d-fdcd8e02c657

🔁 Steps to Reproduce

1. Mac OS
2. Using Claude Code CLI provider, particularly in code mode. Try to provide any command at all.

💥 Outcome Summary

1. Expected a proper, parsed response, and clean file edits, but got:

• an unparsed json response from the assistant message, with a bunch of /n/n/n entries littered throughout. Example:

"{"type":"assistant","message":{"id":"msg_01BAY8woVPxXK7eAF5UCm8ov","type":"message","role":"assistant","model":"claude-sonnet-4-20250514","content":[{"type":"text","text":"Now I'll research the enhanced TimeoutHandler architecture. Based on the current implementation and PRD requirements, I need to design a comprehensive enhanced version. Let me start implementing the enhanced TimeoutHandler class.\n\n"

• improper file edits with \n\n throughout and often first on a single line rather than properly formatted.

2. Sometimes, I will receive API Timeout errors, or that Claude Code is exceeding the maximum output tokens.

Before the recent updates (in the last day) I would just get "API Request Cancelled". This has been fixed by now providing the actual error, which then is Timeout or exceeded maximum output tokens.

--

Strangely, this seems to ONLY happen in Code Mode.

📄 Relevant Logs or Errors (Optional)

[roomote]

Hi @bpeterson1991! I'm here to help out the maintainers and am going to see if I can fix this issue. I'll investigate the API parsing and response formatting issues you're experiencing with Claude Code CLI in code mode. Thanks for providing the detailed reproduction steps and examples!

[roomote]

Hi @bpeterson1991! I've created a pull request to fix the JSON parsing issue you reported: #6126

The fix includes:

• Detection and parsing of string chunks that contain JSON assistant messages
• Proper extraction and formatting of text content from parsed messages
• Graceful handling of malformed JSON (displays as-is when parsing fails)
• Comprehensive test coverage for the new behavior

All CI checks have passed. This should resolve the issue where unparsed JSON with escaped newlines was being displayed in code mode. The assistant responses should now be properly formatted.

[bpeterson1991]

The other issue not addressed here is the fact that when this starts, then when it tries to edit files it'll create a bunch of files that look like:

"table_config.get('service_name', 'unknown')\n \n ttl = int(time.time()) + (30 * 24 * 60 * 60) # 30 days\n \n self.dynamodb.put_item(\n TableName=self.table_name,\n Item={\n 'handoff_id': {'S': handoff_id},\n 'source_execution_id': {'S': source_execution_id},\n 'target_execution_id': {'S': target_execution_id},\n 'source_lambda': {'S': payload.predecessor_function_name},\n 'target_lambda': {'S': target_lambda},\n 'tenant_id': {'S': tenant_id},\n 'endpoint_key': {'S': f"{service_name}#{table_name}"},\n 'payload_size': {'N': str(payload.compressed_size or payload.original_size)},\n 'handoff_status': {'S': HandoffStatus.INITIATED.value},\n 'initiated_at': {'S': get_timestamp()},\n 'retry_count': {'N': '0'},\n 'ttl': {'N': str(ttl)}\n }\n )\n \n self.logger.info(f"Handoff initiated: {handoff_id}",\n handoff_id=handoff_id,\n source_execution_id=source_execution_id,\n target_execution_id=target_execution_id,\n target_lambda=target_lambda)\n \n return handoff_id\n \n except Exception as e:\n self.logger.error(f"Fa"

[bpeterson1991]

I believe this is connected, only starting on Monday (when I saw all of these issues) am I getting this error:

API Error: Claude's response exceeded the 8000 output token maximum. To configure this behavior, set the CLAUDE_CODE_MAX_OUTPUT_TOKENS environment variable.

I had no issue previously with a maximum of 8000 tokens, so this makes me thing something else has caused the change.

[bpeterson1991]

Further details:

I opened up developer tools and getting this error whenever I see that unparsed json. I'm using RooCode Nightly now to get the latest updates but the same error is happening in RooCode.

[Extension Host] Error parsing chunk: SyntaxError: Unterminated string in JSON at position 10000 (line 1 column 10001)
at JSON.parse ()
at attemptParseChunk (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-code-nightly-0.0.6310/dist/extension.js:653767:17)
at parseChunk (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-code-nightly-0.0.6310/dist/extension.js:653759:18)
at runClaudeCode (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-code-nightly-0.0.6310/dist/extension.js:653642:24)
at process.processTicksAndRejections (node:internal/process/task_queues:105:5)
at async ClaudeCodeHandler.createMessage (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-code-nightly-0.0.6310/dist/extension.js:653826:22)
at async _Task.attemptApiRequest (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-code-nightly-0.0.6310/dist/extension.js:728596:26)
at async _Task.recursivelyMakeClineRequests (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-code-nightly-0.0.6310/dist/extension.js:728313:26)
at async _Task.recursivelyMakeClineRequests (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-code-nightly-0.0.6310/dist/extension.js:728410:31)
at async _Task.initiateTaskLoop (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-code-nightly-0.0.6310/dist/extension.js:728175:26)
at async _Task.resumeTaskFromHistory (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-code-nightly-0.0.6310/dist/extension.js:728092:5) 10000

[bpeterson1991]

Once I expand it to 64000 tokens, I don't get the error above anymore, I'm now getting a rate limit error:

[Extension Host] Rate limit hit, retrying in 5000ms (attempt 2/3)
console.ts:137 [Extension Host] Rate limit hit, retrying in 10000ms (attempt 2/3)
console.ts:137 [Extension Host] Rate limit hit, retrying in 20000ms (attempt 2/3)
console.ts:137 [Extension Host] Rate limit hit, retrying in 40000ms (attempt 2/3)
console.ts:137 [Extension Host] Rate limit hit, retrying in 80000ms (attempt 2/3)

As well as:

[DirectoryScanner] Error processing batch (attempt 1) in workspace /Users/brandonpeterson/Desktop/avesa: Error: Failed to create embeddings after 3 attempts: HTTP 429 - 429 Resource has been exhausted (e.g. check quota).
at W1e (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-cline-3.23.18/dist/extension.js:1069:5474)
at t._embedBatchWithRetries (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-cline-3.23.18/dist/extension.js:1069:18620)
at process.processTicksAndRejections (node:internal/process/task_queues:105:5)
at async t.createEmbeddings (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-cline-3.23.18/dist/extension.js:1069:16567)
at async t.createEmbeddings (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-cline-3.23.18/dist/extension.js:1069:20749)
at async iwe.processBatch (/Users/brandonpeterson/.vscode/extensions/rooveterinaryinc.roo-cline-3.23.18/dist/extension.js:2742:7055)