This release includes breaking changes for platform teams planning a safe upgrade.
✓ No known CVEs patched in this version
Topics
Summary
AI summaryUpdates Bug Fixes, code, and https://github.com/langchain-ai/deepagents/issues/4816 across a mixed release.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Feature | Low |
Adds classifier-backed Auto approval mode behind DEEPAGENTS_CODE_EXPERIMENTAL=1 Adds classifier-backed Auto approval mode behind DEEPAGENTS_CODE_EXPERIMENTAL=1 Source: llm_adapter@2026-07-17 Confidence: high |
— |
| Feature | Low |
Adds shutdown toast for deferred exits Adds shutdown toast for deferred exits Source: llm_adapter@2026-07-17 Confidence: high |
— |
| Feature | Low |
Allows expanding truncated task descriptions via click or Ctrl+O Allows expanding truncated task descriptions via click or Ctrl+O Source: llm_adapter@2026-07-17 Confidence: high |
— |
| Feature | Low |
Persists Debug Console clear (Ctrl+L) state after reopening Persists Debug Console clear (Ctrl+L) state after reopening Source: llm_adapter@2026-07-17 Confidence: high |
— |
| Feature | Low |
Adds debug logging for skill-name override collisions Adds debug logging for skill-name override collisions Source: llm_adapter@2026-07-17 Confidence: high |
— |
| Bugfix | Medium |
Keeps chat input responsive during /restart command Keeps chat input responsive during /restart command Source: llm_adapter@2026-07-17 Confidence: high |
— |
| Bugfix | Medium |
Preserves paste placeholders when backspacing a newline below them Preserves paste placeholders when backspacing a newline below them Source: llm_adapter@2026-07-17 Confidence: high |
— |
| Bugfix | Medium |
Makes markdown AppMessage output selectable Makes markdown AppMessage output selectable Source: llm_adapter@2026-07-17 Confidence: high |
— |
| Bugfix | Medium |
Counts only running tools in live tool‑group line display Counts only running tools in live tool‑group line display Source: llm_adapter@2026-07-17 Confidence: high |
— |
| Bugfix | Medium |
Reduces repeated probing of an unreachable Ollama daemon to once per reload Reduces repeated probing of an unreachable Ollama daemon to once per reload Source: llm_adapter@2026-07-17 Confidence: high |
— |
| Bugfix | Medium |
Keeps task timers monotonic across nested subagent human‑in‑the‑loop flows Keeps task timers monotonic across nested subagent human‑in‑the‑loop flows Source: granite4.1:30b@2026-07-17-audit Confidence: low |
— |
| Bugfix | Low |
Preserves goal criteria proposals when marker clearing fails Preserves goal criteria proposals when marker clearing fails Source: granite4.1:30b@2026-07-17-audit Confidence: low |
— |
| Bugfix | Low |
Quiets MCP auth‑skip debug logging for known patterns Quiets MCP auth‑skip debug logging for known patterns Source: granite4.1:30b@2026-07-17-audit Confidence: low |
— |
| Bugfix | Low |
Improves /version diagnostics for editable installs and core dependency reporting, surfacing langchain-quickjs Improves /version diagnostics for editable installs and core dependency reporting, surfacing langchain-quickjs Source: granite4.1:30b@2026-07-17-audit Confidence: low |
— |
| Bugfix | Low |
Removes the uv install tip from the /version update hint Removes the uv install tip from the /version update hint Source: granite4.1:30b@2026-07-17-audit Confidence: low |
— |
Full changelog
Features
- Added classifier-backed Auto approval mode behind
DEEPAGENTS_CODE_EXPERIMENTAL=1(#4804). - Added a shutdown toast for deferred exits (#4830).
- Task descriptions that were truncated can now be expanded by clicking or pressing
Ctrl+O(#4811). - Debug Console clears with
Ctrl+Lnow persist after reopening (#4812). - Added debug logging for skill-name override collisions (#4772).
Bug Fixes
- Keep chat input responsive during
/restart(#4808). - Fixed paste placeholders disappearing when backspacing a newline below them (#4757).
- Made markdown
AppMessageoutput selectable (#4814). - Fixed live tool-group counts to include only running tools (#4809).
- Kept
tasktimers monotonic across nested subagent human-in-the-loop flows (#4771). - Preserved goal criteria proposals when marker clearing fails (#4785).
- Reduced repeated probing of an unreachable Ollama daemon to once per reload (#4806).
- Quieted MCP auth-skip debug logging for known patterns (#4805).
- Improved
/versiondiagnostics for editable installs and core dependency reporting, including surfacinglangchain-quickjs(#4816, #4813). - Removed the
uv installtip from the/versionupdate hint (#4822).
Internal maintainers: @mdrxy
Released by: @mdrxy
Git log since deepagents-code==0.1.42This commit history includes changes to this package. Commits are listed newest first.
e14e0adrelease(deepagents-code): 0.1.43 (#4825)eae7c28feat(code): classifier-backed Auto approval mode (#4804)a8a05f5feat(code): shutdown toast for deferred exits (#4830)3f82114refactor(code): dropconfig list/lsaliases forconfig show(#4783)a915c59fix(code): keep chat input responsive during/restart(#4808)7408378fix(code): disentangle version diagnostics for editable installs (#4816)6952011fix(code): keeptasktimers monotonic across nested subagent HITL (#4771)4e618f0fix(code): dropuv installtip from/versionupdate hint (#4822)487b25ffix(code): keep goal criteria proposal when marker clear fails (#4785)c2c27d6fix(code): quiet MCP auth-skip debug logging for known patterns (#4805)2ddb551feat(code): expand truncatedtaskdescriptions on click orCtrl+O(#4811)f5c3259style(code): show enabled plugin status in green (#4821)ea82538fix(code): keep paste placeholder when backspacing newline below it (#4757)be8e521style(code): drop hardcoded indent fromlsandwrite_todosoutput (#4762)77603c6feat(code): debug-log skill-name override collisions (#4772)45baf6cfix(code): probe unreachable Ollama daemon once per reload (#4806)4d483fdfix(code): count only running tools in live tool-group line (#4809)20089a5feat(code): persist Debug ConsoleCtrl+Lclear across reopen (#4812)2d6b2cefix(code): make markdownAppMessageoutput selectable (#4814)4112392fix(code): surfacelangchain-quickjsin/versioncore deps (#4813)
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About deepagents
Agent harness built with LangChain and LangGraph. Equipped with a planning tool, a filesystem backend, and the ability to spawn subagents - well-equipped to handle complex agentic tasks.
Related context
Related tools
Earlier breaking changes
- vlangchain-quickjs==0.2.0 Removes `skills_backend` capability.
- vdeepagents-cli==0.2.0 `deepagents deploy` now targets Managed Deep Agents API; removes `deepagents dev`, `deepagents.toml`, and `--config`/`--dry-run` flags.
Beta — feedback welcome: [email protected]