0.133.22

What's Changed

0.133.22 - 2026-06-28

Highlights

Bug Fixes

• Stream isolated review subcalls (a7ffd7f) (@kernitus)
• Ignore responses keepalive events (f33650f) (@kernitus)
• Update compatible responses retry test (7b73ba2) (@kernitus)

Documentation

• Add web_search tool reference, document /continue and did-you-mean (90a321b)
• Expand TODO with known tech debt items (144491a)
• Add defuddle_fetch tool reference and web_search config fields (5040fac)
• Fix stale src/tui.rs references, update CONTRIBUTING.md and INDEX.md (8b0518e)
• Audit and trim per-crate AGENTS.md files (018656e)
• Align CI/CD docs with actual workflow files (5e39c04)

Features

• Implement cross-turn tracking and limit assistant text responses to prevent runaway loops (466c0c2)
• Add tiktoken integration for token counting and hierarchical summarization (3d5af4e)

Other Changes

Other

• Convert stringly-typed ResponseId, OutputItemId, ContentPartId to newtypes (2df6001)
• Implement recovery fallback messages and enhance outline search with directory summaries (75429ee)
• Implement few-shot example selection and rendering; enhance tool descriptions and documentation. (82f078b)
• Add persistent audit logging and per-MCP-server sandboxing (086208e)
• Implement async approval blocker and escalation gate for irreversible actions (32d15be)
• Add task history, message metadata, pending actions, and state schema modules (b670e3a)
• Enhance tool descriptions for clarity and guidance on usage (b071d6a)
• Enhance agent execution flow with escalation handling and stagnation detection (209364a)
• Optimize token count management and enhance loop detection efficiency (e1f90ce)
• Refactor escalation decision enum and clean up evaluator gate outcome handling (c27527d)
• Refine system prompt guidelines for clarity and simplicity in agent operations (ae9a8a8)
• Clarify description for 'follow' parameter in unified search schema (e86300c)
• Enhance skill availability messages to include instructions for listing full catalog (5512d9d)
• Refactor tool outcome handling and improve skill overflow messages for clarity (782b8fc)
• Refactor model ID matching and clean up import statements for clarity (764128b)
• Refactor string formatting for improved readability and consistency across tests and core modules (87a375f)

Tests

• Fix prompt cache retention fixtures (6d03201) (@kernitus)

Full Changelog: 0.133.21...0.133.22