# Cline SDK v0.2.1 changes tool-result caps and reasoning handling

published: 2026-11-21T12:05:35.684Z
topic: frameworks
source: https://github.com/cline/cline/releases/tag/sdk%2Fsdk%2Fv0.2.1
url: https://aiagentnews.io/stories/cline-sdk-v0-2-1-changes-tool-result-caps-and-reasoning-handling

Cline's SDK v0.2.1 sends read_files results to providers under a 50k cap, while other tool results keep an 8k limit. The release also fixes reasoning-related request errors for some models and refreshes the catalog to 7,621 models.

The release changes how tool results are capped. read_files output now uses a 50k file-content cap instead of the 8k tool-result cap, so a file the tool reports in full reaches the provider in full. Command, search, MCP and custom tool results keep the 8k cap.

Several reasoning fixes ship in this version. Models that need interleaved reasoning now receive a placeholder reasoning_content on assistant messages that lack one, which addresses a 400 error Reflection Beam returned after compaction or a model switch. The native OpenAI adapter also fits the reasoning level to each model's advertised effort ladder, so max and Extra High are no longer sent to GPT-5.x and o-series models that reject them.

The default model for the cline provider is now the first bundled free model, and CLINE_DEFAULT_MODEL_ID is set to anthropic/claude-sonnet-5.5. The model catalog grows from 6,457 to 7,621 models. The release also removes CLINE_CAPTURE_* provider request capture, which wrote full request bodies to disk.

## Why it matters if you build agents

Agents that read large files through Cline no longer get those reads cut at 8k, so prompts can carry more complete file contents. Teams running reasoning models on OpenAI-compatible endpoints should confirm their requests still work under the new reasoning handling. Any logging that depended on CLINE_CAPTURE_* needs another approach.
