Fixed prompt and agent hooks handling for instruction-based blocking
Fixed a bug where prompt and agent hooks written as blocking instructions were not working correctly. Improved evaluation of Stop and SubagentStop hooks so Claude is less likely to stop early when hooks are written as conditional instructions.
- Fixed
promptandagenthooks written as instructions (such as "Block commands that...") allowing what they should block - Improved how
prompthooks on Stop and SubagentStop written as instructions (such as "Carry on if the build is broken") are judged, so Claude is less likely to stop early
Source: original entry ↗
More from Claude Code
Follow Claude Code to get its new changes in your feed and email digest.
Claude Code 2.1.293
Released version 2.1.293 with Claude Haiku 5.5 as the default Haiku model featuring 1M context window, expanded agent functionality with agentType and isDeferred fields, and numerous bug fixes across context management, MCP connections, vim mode, permissions, and cloud sessions.
Claude Code 2.1.292 Release
Claude Code 2.1.292 includes new features like marketplace source support for plugin installation, effort parameters for agent tools, and prompt caching for mods. This release also addresses numerous security fixes including permission bypass issues, file access restrictions, and cache tampering vulnerabilities, plus improvements to session handling, auto mode restoration, and various UI/UX fixes.
Fixed regressions with cloud sessions and message handling
Fixed a regression where cloud sessions could drop answers to permission prompts and another regression where the last messages of a session could be lost when quitting.