Automated Release PR

0.11.5 (2026-05-29)

Full Changelog: v0.11.4...v0.11.5

Features

• api: add cleaned_at field to task response types (38ed338)
• deps: bump openai-agents to >=0.14.3 for scale-sandbox oai_agents adapter (#375) (e1b31d9)

Performance Improvements

• tracing: span queue linger + per-loop httpx keepalive (#362) (feec842)

Chores

• deps: drop unused runtime deps and exclude tests from wheel (#367) (f4303d1)

Refactors

• types: promote protocol types to agentex.protocol.* (#371) (6f1c14f)

This pull request is managed by Stainless's GitHub App.

The semver version number is based on included commit messages. Alternatively, you can manually set the version number in the title of this pull request.

For a better experience, it is recommended to use either rebase-merge or squash-merge when merging this pull request.

🔗 Stainless website
📚 Read the docs
🙋 Reach out for help or questions

Greptile Summary

This is the 0.11.5 release PR, bundling four changesets: tracing performance improvements, a protocol-type module promotion, new cleaned_at fields on task response types, and dependency cleanup.

• Tracing perf (perf(tracing): span queue linger + per-loop httpx keepalive #362): AsyncSpanQueue gains a configurable linger window (default 100 ms) to coalesce spans into larger batches, per-event-loop HTTP client caching via WeakKeyDictionary to restore connection keepalive safely, bounded retry on transient 429/5xx failures, and a dropped_spans counter for observability.
• Protocol types (refactor(types): promote protocol types to agentex.protocol.* #371): Wire-protocol shapes (RPCMethod, CreateTaskParams, JSONRPCRequest, etc.) are promoted from agentex.lib.types.* to the new canonical agentex.protocol.* package; the old paths become re-export shims with identity-preserving tests to guard isinstance compatibility.
• API surface (#38ed338): cleaned_at: Optional[datetime] added to Task, TaskListResponseItem, TaskRetrieveResponse, and TaskRetrieveByNameResponse.
• Deps (chore(deps): drop unused runtime deps and exclude tests from wheel #367): Several test/dev-only packages (pytest, pytest-asyncio, ipykernel, datadog, anthropic, tzdata, tornado) removed from the runtime wheel; test directories excluded from the built wheel via hatch config.

Confidence Score: 5/5

Safe to merge — changes are well-scoped, each guarded by matching tests, and backward compatibility is explicitly validated.

All three major areas of change (span-queue drain overhaul, per-loop client caching, protocol-type promotion) are covered by new or updated unit tests that pin the exact contracts being changed. The retry/drop logic is internally consistent, the WeakKeyDictionary eviction is verified end-to-end, and the back-compat shims are identity-tested. Dependency removals are dev/test packages that were never appropriate in the runtime wheel.

No files require special attention.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    E[enqueue span] --> S{_stopping?}
    S -- yes --> D1[record_drop: shutting down]
    S -- no --> Q{queue full?}
    Q -- yes --> D2[record_drop: queue full]
    Q -- no --> QI[put item on asyncio.Queue\nattempts=0]

    QI --> DRAIN[drain loop: await queue.get]
    DRAIN --> FIRST[batch = first item]
    FIRST --> LINGER{linger_ms > 0\nand not stopping?}
    LINGER -- yes --> WAIT[wait_for queue.get\nup to linger deadline]
    WAIT --> MORE{more items\nor timeout?}
    MORE -- more items --> BATCH_FULL{batch == batch_size?}
    BATCH_FULL -- no --> WAIT
    BATCH_FULL -- yes --> SPLIT
    MORE -- timeout --> SPLIT[split START / END]
    LINGER -- no --> DRAIN_NOW[drain_nowait until\nbatch_size or QueueEmpty]
    DRAIN_NOW --> SPLIT

    SPLIT --> PROC[_process_items per event_type\ngather per-processor _handle]
    PROC --> TRY{processor call\nsucceeds?}
    TRY -- yes --> CLEAR[batch.clear loop back]
    CLEAR --> DRAIN
    TRY -- no --> RETRY{retryable status\n429/5xx?}
    RETRY -- no --> D3[record_drop: permanent failure]
    RETRY -- yes --> ATCHECK{attempts+1 < max_retries?}
    ATCHECK -- no --> D4[record_drop: retries exhausted]
    ATCHECK -- yes --> REENQUEUE[_reenqueue\nattempts+1, processors=p]
    REENQUEUE --> Q2{queue full?}
    Q2 -- yes --> D5[record_drop: queue full on retry]
    Q2 -- no --> DRAIN

    subgraph PerLoopCache [Per-event-loop client cache]
        LC[asyncio.get_running_loop] --> WKD[WeakKeyDictionary lookup]
        WKD -- miss --> BUILD[_build_client keepalive=20]
        BUILD --> STORE[store in WeakKeyDictionary]
        WKD -- hit --> USE[reuse existing client]
    end

    PROC -.->|HTTP call via| PerLoopCache

Reviews (6): Last reviewed commit: "release: 0.11.5" | Re-trigger Greptile