v0.19.1
AgentCrew v0.19.1 Release Changelog
Release date: 2026-07-22
✨ Features
- New models – Added support for
inkling(Together),gemini-3.5-flash-lite/gemini-3.6-flash(Google), and fixed provider/service routing for GPT-5.6 models (provider="openai",service_name="openai_response"). - Code analysis caching –
analyze_reporesults are now cached locally (under.agentcrew/analyze_repo_cache/) with Git‑based invalidation and LRU eviction. Repeated calls return instantly. - Symbol lookup tools – Two new tools
find_definitionandfind_referencesallow agents to quickly locate symbol declarations and usages using Tree‑sitter syntax (Python, JavaScript, and more). /copycommand – Re‑introduced/copy <num>to copy the nth‑latest assistant response to the clipboard. Works in both console (Ctrl+Shift+C) and GUI. Context‑aware tab‑completion available.- Prompt evolution feedback – When running
/evolve, users can now optionally answer three guiding questions about what they want to change, what the agent does well, and what it does poorly. This feedback is incorporated into the analysis prompt.
🐛 Bug Fixes
- Search mode in conversation browser – Fixed a regression where keyboard bindings (
k,j,l) consumed characters in search mode, preventing proper typing. create-agentcrash – Fixed missingmodel_idparameter and removed stalememory_pathhandling that caused command to fail.- OpenAI model provider – Corrected
providerfor GPT-5.6 models (changed fromopenai_responsetoopenaiwith aservice_namesub‑field). - Console flicker – Resolved screen flickering during fast‑streaming output by enabling
auto_refreshin the live update panel.
📚 Documentation
- Updated
PLUGIN_DEVELOPMENT.md– clarified trust boundary configuration (--trusted-project-plugins,config.json), line‑wrap improvements, and example fix.
🔧 Chores
- Updated
tree-sitter-language-packdependency from1.6.2→1.12.5(better language coverage). - Updated CrofAI vision model from
qwen3.6-27b→gemma-4-31b-it.
Thank you for using AgentCrew! If you encounter any issues, please open a GitHub issue.