Preserve newline before leaked CONTENT_FILTER and make XML tool-call parsing more robust

[CJackHwang]

Motivation

• Prevent dropping legitimate line breaks when upstream models append leaked CONTENT_FILTER markers after a newline.
• Improve resilience of XML-style tool call parsing to non-XML-safe content (for example raw & in shell commands) and better extract tool names and parameters from varied tag names.

Description

• Update filterLeakedContentFilterParts/stripLeakedContentFilterSuffix behavior to preserve trailing newline characters and factor the drop logic into shouldDropCleanedLeakedChunk which keeps newline-only chunks.
• Add shouldDropCleanedLeakedChunk to avoid dropping chunks that contain \n while still removing empty/whitespace-only chunks.
• Add regex-based extraction for tool name tags via xmlToolCallNameTagNames and xmlToolCallNamePatternByTag and attempt tag-based parameter parsing before full XML decoding to handle non-XML-safe payloads.
• Enhance parameter parsing in parseSingleXMLToolCall to try parseXMLChildKV when parseToolCallInput yields only _raw, and to merge parsed key/value pairs into params when present.
• Add unit tests in internal/sse/line_test.go and internal/util/toolcalls_test.go covering preserved newlines before leaked CONTENT_FILTER and XML parameter JSON containing raw ampersands.

Testing

• Ran go test ./... including SSE and util packages and all tests completed successfully.
• New tests TestParseDeepSeekContentLinePreservesTrailingNewlineBeforeLeakedContentFilter and TestParseDeepSeekContentLineKeepsNewlineOnlyChunkBeforeLeakedContentFilter passed.
• New test TestParseToolCallsSupportsXMLParametersJSONWithRawAmpersand passed and verifies preservation of raw shell commands in parameters.

---

Codex Task

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
ds2api	Ready	Preview, Comment	Apr 3, 2026 5:46pm

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6878d81a75

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Defer early return until all XML arg tags are considered

This early return makes parseSingleXMLToolCall commit to a single argument block chosen by findMarkupTagValue priority, so mixed payloads like <parameters>{...}</parameters><arguments>{...}</arguments> now lose fields that the XML decoder previously merged. In that case the parser returns from the regex path with only one map and never reaches the decoder loop, which can change tool inputs and break calls that rely on keys from the skipped tag.

Useful? React with 👍 / 👎.