This release adds 3 notable features for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
Summary
AI summaryAdded AskUser tool, Hooks API, Session.steer() and interrupt(), and hardened abort handling.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Feature | Medium |
Adds AskUser tool for structured user input during agent runs Adds AskUser tool for structured user input during agent runs Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Feature | Medium |
Introduces Hooks system with PreToolUse, PostToolUse, UserPromptSubmit, Stop, and PreCompact APIs Introduces Hooks system with PreToolUse, PostToolUse, UserPromptSubmit, Stop, and PreCompact APIs Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Feature | Medium |
Adds Session.steer() and Session.interrupt() for run control without aborting state Adds Session.steer() and Session.interrupt() for run control without aborting state Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Performance | Low |
Improves adjacent-batch partitioning in tool scheduler to preserve strict result ordering Improves adjacent-batch partitioning in tool scheduler to preserve strict result ordering Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Bugfix | Medium |
Hardens abort signal propagation across parallel tool execution, emitting tool_call_end before AbortError Hardens abort signal propagation across parallel tool execution, emitting tool_call_end before AbortError Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Bugfix | Medium |
Strengthens session store concurrency guarantees for updateMeta calls without patch loss Strengthens session store concurrency guarantees for updateMeta calls without patch loss Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Bugfix | Medium |
Expands provider error normalization for OpenAI APIs, handling retry-after headers and mid-stream aborts Expands provider error normalization for OpenAI APIs, handling retry-after headers and mid-stream aborts Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Bugfix | Medium |
Refines compaction logic to re-inject skill listings and invoked skill bodies after context threshold Refines compaction logic to re-inject skill listings and invoked skill bodies after context threshold Source: llm_adapter@2026-06-12 Confidence: high |
— |
Full changelog
What's Changed
Added
- AskUser tool — Agents can now pause mid-run to elicit structured input from the user. Supports single-select, multi-select, and free-text responses with a 1–4 question format, option validation, and graceful decline handling.
- Hooks system — Introduced a first-class hook API (
PreToolUse,PostToolUse,UserPromptSubmit,Stop,PreCompact) that allows consumers to intercept, modify, or block agent actions at runtime without modifying core loop logic. - Steering & interruption — Added
Session.steer()andSession.interrupt()for active-run control.steer()injects a user message at the next turn boundary without aborting the run;interrupt()ends the current run cleanly while preserving all in-flight turn state.
Improved
- Hardened abort signal propagation across parallel tool execution, ensuring in-flight tool calls emit proper
tool_call_endevents before surfacingAbortError. - Improved adjacent-batch partitioning in the tool scheduler to preserve strict result ordering across mixed read/write batches.
- Strengthened session store concurrency guarantees; concurrent
updateMetacalls now correctly accumulate all patches without loss. - Expanded provider error normalization for OpenAI Chat and Responses APIs, covering HTTP-date
retry-afterheaders and mid-stream abort detection. - Refined compaction logic to correctly re-inject skill listings and invoked skill bodies after context threshold is crossed.
Full Changelog: https://github.com/skawld/skawld-sdk/blob/main/CHANGELOG.md
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 Skawld
All releases →Beta — feedback welcome: [email protected]