record-agent-failurelisted

# Record Agent Failure Do not trigger for an ordinary new requirement or refinement. Trigger only when feedback challenges an earlier outcome, reports recurrence, or explicitly asks for failure qualification/learning. Then make one fast classification from evidence that existed before the outcome: - `requirement_update`, `requirement_clarification`, or `preference_update`: not a failure. - `real_failure`: a prior invariant, inspectable mismatch, material impact or recurrence risk, controllable cause, and durable prevention are all evidenced. - `mixed`: separate only the prior-invariant failure into `failure_portion`; keep new work out of memory. - `uncertain`: chronology or required evidence cannot be established. Once triggered, normally call `remember_failure` once even when the classification is a non-failure or `uncertain`; this preserves false-positive qualification telemetry. Those classifications use only: ```json {"summary":"compact chronology","classification":"requirement_update"} ``` For `mixed`, `failure_portion` is a compact string containing only the old-invariant mismatch. Real or mixed failures use the exact tool schema: `expectation` has `invariant`, `source`, `evidence`; `observed` has `outcome`, `impact`, optional `recurrence_risk`; `cause` has `layer`, `failure_mode`, `component`, `evidence`, `recommended_change`, and `verification`; `lesson` has `rule`, `prevention`, `verification`, and optional `title`, `applicability`, `counterexamples`. Use only