[leehack]

Summary

• Qwen3-Coder and Ling 3 tool-call grammars: pick a -at-value-end copy name that no converter rule holds, so a nested property named at-value-end no longer defines a rule twice and rejects valid calls (0.19.x audit follow-ups (index) #1861, Parse Qwen3-Coder XML call bodies as llama.cpp does (#1898) #1906 re-audit).
• Seed-OSS: when a closed valid call is followed by two unclosed scopes, the added </seed:tool_call> no longer ends the returned content; the calls are kept (0.19.x audit follow-ups (index) #1861, Parse Seed-OSS tool calls closer to llama.cpp (#1898) #1908 re-audit).

The grammar change only renames rules; llama.cpp errors on the Seed-OSS input, so there is no new oracle row.

Production-readiness scope

• User-facing scope: Valid Qwen3-Coder/Ling 3 calls are accepted for such schemas; Seed-OSS content is clean.
• Supported platforms/paths: llama.cpp-template parsing and grammar (VM).
• Unsupported or intentionally unavailable paths: None changed.
• Out of scope / follow-ups: Remaining 0.19.x audit follow-ups (index) #1861 items are tracked there.

Completeness checklist

• Declared scope is fully implemented.
• Unsupported platform/option combinations: unchanged.
• Changelog entries updated where user-visible (N/A for tooling-only).
• Regression coverage covers the original finding and fails first against origin/main.
• Validation coverage registry: N/A, no capability change.
• No secrets or signed URLs leak through logs or errors.
• Follow-ups: None beyond 0.19.x audit follow-ups (index) #1861.

PR type guidance

• Bugfix PR: root cause and regression test are in the summary.

Test Plan

• dart run tool/prepare_workspace.dart
• dart format --output=none --set-exit-if-changed .
• dart analyze
• Targeted VM: xml_argument_grammar_test, seed_oss_handler_test (each new test fails against origin/main code)
• run_template_parity_suites.sh, coverage-lib: results in a PR comment

Matrix Evidence

Matrix row	Scope covered	Platform / model / backend	Result	Evidence / notes
structured-output-adversarial	parser/grammar change	VM	pending	see comment
root-vm	targeted suites	macOS VM	PASS
root-chrome	N/A	no web code touched	N/A

Review Notes

• Independent review status: N/A (not yet requested)
• CI status / head SHA: pending

High-risk regression review

• Classification: high-risk (structuredOutput)
• Implementation task: lane-1861 (0.19.x audit follow-ups (index) #1861)
• Independent blocking QA task: pending (draft; not requested)
• Exact head / current base: 835d130 / edaf8b5
• Production-branch deletion, bypass, or miswire proof: reverting each lib change fails its new test
• Affected-family real-model/artifact evidence: N/A: no model-dependent change; unit tests plus llama.cpp cd959bae9 template parity suites
• Explicit unavailable-family or other N/A evidence: N/A
• Known PR-caused P1 regressions: 0
• Unresolved review threads: 0

[leehack]

Landed via batch integration PR #1926 (rebase merge). Integration audit PASS at 6439022; this PR's audited head is recorded in #1926's constituent table.

[github-actions]

Chat app preview removed for leehack/llamadart-chat-pr-1919.