fix: harden protocol handling and rate-limit recovery - #269
Open
angelarose210 wants to merge 5 commits into
Open
fix: harden protocol handling and rate-limit recovery#269angelarose210 wants to merge 5 commits into
angelarose210 wants to merge 5 commits into
Conversation
- Downgrade unknown content block types to text instead of 422 errors - Accept free-form message roles (Claude Code sends non-standard roles) - Add first-token timeout to prevent indefinite hangs on stalled upstream - Improve truncation recovery with distinct small/transient vs large/size messaging - Add usage logger for dashboard tracking Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…tics Improve robustness for GPT models proxied through the gateway while keeping Claude request/response behavior byte-identical: - parsers: classify malformed/truncated tool-call arguments as non-executable protocol failures and record metadata-only anomalies - streaming_core: warning-only detection of raw transport-text leaks in content (never mutates the response) - debug_logger: add record_anomaly()/anomaly_info.json, gated behind DEBUG_MODE (off by default); persisted on both success and error paths, scalars only - protocol_anomalies: new pure classifier for high-confidence transport leaks - carry forward prior-session converter/streaming/truncation-recovery changes - update stale tests to match intcentional role-normalization and the small/large truncation message split; document anomaly_info.json in .env.example - gitignore .venv/ Full suite: 1728 passed.
… path collect_stream_response() flattened every tool_calls delta with `tool_calls.extend(...)`, ignoring the `index` field. That is correct only when a provider emits each tool call in exactly one chunk. Per the OpenAI streaming protocol a tool call may be split — identity (id + function.name) in one delta, argument fragments in later deltas keyed by the same index — so any split emission produced multiple half-formed tool calls: one with a name and no arguments, another with arguments and no name. Accumulate by index the way a streaming client does (see the Vercel AI SDK StreamingToolCallTracker), taking identity from the first delta for an index and concatenating argument fragments, then emit in index order. Latent today: the gateway currently emits one chunk per tool call, verified end to end against a live gateway (direct curl, tool-result round trip, and parallel multi-tool turns). This hardens the path against split emission rather than fixing an observed failure. Tests: reassembly for single and parallel tool calls, non-streaming collection, and the existing None-protection cases retargeted at the reassembled call instead of individual fragments.
|
Thanks for the PR! 🎉 Before merge, we need a one-time CLA confirmation. Full CLA text: Please reply once with: You need to write once, all further messages from me can be ignored. |
Author
|
I have read the CLA and I accept its terms |
|
Thanks for the PR! 🎉 Before merge, we need a one-time CLA confirmation. Full CLA text: Please reply once with: You need to write once, all further messages from me can be ignored. |
|
@angelarose210 could you give a try to this gateway: https://github.com/ankitcharolia/kiro-gateway It works quite well with All AI harness and actively being developed. The most important thing is that it is ACP compliant |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary