2 Commits

Author	SHA1	Message	Date
chengyongru	0a396aa6e2	Improve tool call validation strictness (#4190) ... * Improve tool call validation strictness Reject near-miss tool names without executing suggested tools. Require object-shaped tool parameters while preserving only lossless JSON wire-shape normalization. * Tighten tool call argument validation * Simplify tool argument validation tests * Improve tool name suggestions * Simplify tool suggestion helpers * Limit tool suggestions to canonical matches * Allow repair only for tool history replay * Clarify non-object tool argument errors * Inline replay tool argument normalization * Track only successful tool executions * Reject JSON null tool arguments	2026-06-09 14:50:40 +08:00
chengyongru	99cc6ee808	test(agent): expand coverage and refactor test structure ... - Add 42 tests for ContextBuilder (context.py: 0→42 tests) - Add 37 tests for SubagentManager lifecycle (subagent.py: 2→37 tests) - Add 42 unit tests for AutoCompact in isolation - Split monolithic test_runner.py (3313 lines) into 9 focused files: test_runner_core, test_runner_hooks, test_runner_errors, test_runner_safety, test_runner_persistence, test_runner_governance, test_runner_tool_execution, test_runner_injections, test_loop_runner_integration - Add 3 config passthrough tests (temperature/max_tokens/reasoning_effort) - Fix fragile patch.object(__init__) in test_stop_preserves_context - Create shared conftest.py with make_provider/make_loop factories Total: 934 tests passing, 0 regressions	2026-05-13 12:49:17 +08:00