SDK hosts get machinePolicy.didNotLoad and matchDeniedModels() for policy handling, plus a tool_progress heartbeat during long hooks. Compaction forking and MCP progress fixes remove bugs in session handling.
Key takeaways · AI-distilled
In Claude AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → SDK 0.3.323, if a tool's permission check fails with an error after a PreToolUse agent hookA script the harness runs automatically at a fixed point in an agent's loop — before a tool call, after an edit, on session start.Full definition → allowed the call, the call now goes to canUseTool, or is refused when no canUseTool callback is set.
The release stops canUseTool from suggesting a Bash allow rule for commands that write to sensitive paths such as .git or .claude; the notes say such a rule did not stop the next request.
resolveSettings() no longer keeps its first read of a macOS MDM profile or Windows registry policy for the whole process: it re-reads after 30 minutes, sooner after a failed read. A new alpha field, machinePolicy.didNotLoad, flags a policy source that may exist but did not load.
While a turn waits on long-running hooks, the SDK now sends a periodic tool_progress heartbeat. Its tool_name is the hook event and its tool_use_id begins with "hook-", and parent_tool_use_id may name no tool call.
forkSession() with upToMessageId now handles context compactionSummarizing an agent's earlier conversation to free room in the context window so a long session can keep going.Full definition → correctly: forking at a compaction summary no longer brings back messages kept after it, and forking at a compaction boundary keeps the conversation before it.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
agent hook — A script the harness runs automatically at a fixed point in an agent's loop — before a tool call, after an edit, on session start.
multi-agent — Using several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.
context compaction — Summarizing an agent's earlier conversation to free room in the context window so a long session can keep going.
