This release adds 3 notable features for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
Summary
AI summaryEscape key now cancels processing loops and dynamic command tools return both stdout and stderr.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Feature | Medium |
Select nested model profiles with `--model <profile>` or config keys, inheriting parent settings Select nested model profiles with `--model <profile>` or config keys, inheriting parent settings Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Feature | Medium |
Allow instructions after multi-line opening delimiter or same‑line enclosed in matching delimiters Allow instructions after multi-line opening delimiter or same‑line enclosed in matching delimiters Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Bugfix | Medium |
Revert incomplete edits on stream errors instead of continuing marker cleanup Revert incomplete edits on stream errors instead of continuing marker cleanup Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Bugfix | Medium |
Reliably remove complete multi‑line marker spans, including decorated closing delimiters Reliably remove complete multi‑line marker spans, including decorated closing delimiters Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Bugfix | Medium |
Preserve task markers and report an error when no substantive edit is made Preserve task markers and report an error when no substantive edit is made Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Bugfix | Medium |
Improve runtime tool‑call output and show diffs for each successful `edit_file` call Improve runtime tool‑call output and show diffs for each successful `edit_file` call Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Bugfix | Medium |
Avoid returning already‑known `read_file` lines until the file changes or an edit resets the task's read history Avoid returning already‑known `read_file` lines until the file changes or an edit resets the task's read history Source: llm_adapter@2026-06-11 Confidence: low |
— |
| Bugfix | Medium |
Use Escape instead of Space to cancel the current processing loop and report `[user cancel received]` Use Escape instead of Space to cancel the current processing loop and report `[user cancel received]` Source: llm_adapter@2026-06-11 Confidence: low |
— |
| Bugfix | Medium |
Name dynamic command tools `D<N>` after their declaration line number Name dynamic command tools `D<N>` after their declaration line number Source: llm_adapter@2026-06-11 Confidence: low |
— |
| Bugfix | Medium |
Return both stdout and stderr from dynamic command tools Return both stdout and stderr from dynamic command tools Source: llm_adapter@2026-06-11 Confidence: low |
— |
Full changelog
Added
- Select nested model profiles with
--model <profile>or the top-leveldefault_model/default-modelconfig key, with inherited parent settings - Put an instruction after a multi-line opening delimiter or enclose a same-line instruction in matching delimiters
Changed
- Use Escape instead of Space to cancel the current processing loop and report
[user cancel received] - Name dynamic command tools
D<N>after their declaration line number - Return both stdout and stderr from dynamic command tools
- Avoid returning already-known
read_filelines until the file changes or an edit resets the task's read history - Improve runtime tool-call output and show diffs for each successful
edit_filecall
Fixed
- Revert incomplete edits on stream errors instead of continuing marker cleanup
- Reliably remove complete multi-line marker spans, including decorated closing delimiters
- Preserve task markers and report an error when the agent makes no substantive edit
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 rik
All releases →Beta — feedback welcome: [email protected]