v3.9.24: post-task learning chain restored (#449)

What's New

Fixes a silent regression in the post-task self-learning chain. aqe hooks post-task was skipping the entire Stream B/D/F pipeline (persistTaskOutcome, updateRoutingOutcomeQuality, updateHookRouterQValue) whenever --task-id was missing — which is the steady state for production hook invocations, because Claude Code does not populate $TOOL_RESULT_agent_id for Task/Agent matchers.

After this fix, every post-task invocation populates rl_q_values, captured_experiences, and routing_outcomes regardless of whether the caller supplied a task id. The Q-learning router can finally accumulate signal.

The same gate was fixed once before in v3.9.21 (patch 030) and quietly came back in v3.9.23. A regression test now guards the no---task-id path so it can't return for a third time.

Thanks to @Jordi-Izquierdo-DDS for the diagnosis, evidence, and patch shape.

Getting Started

npx agentic-qe init --auto

See CHANGELOG and release notes for details.