Tool timeouts can leave AI agents with ambiguous outcomes and partial execution: an action may fail, be committed, or remain in flight while later workflow steps have already finished. We formulate evidence‑guided task recovery over approved obligations, provider capabilities and states compatible with public evidence. A shared admissibility check handles key expiry, authorization timing, dependencies and residual effects; a terminal certificate separates actual goal satisfaction from publicly supported safe completion. The known‑model implementation combines standard observation‑contingent planning with a separately implemented effect‑history adjudicator.\nIn a finite study of 37 assignments, the planning approach and three active controls each certified 28 completions; four saved synthetic fee units required extra computation. A preregistered assignment‑held‑out comparison then evaluated planning advice against evidence prompting using two local model bundles.\nAcross 58 paired units in 19 previously developed task families, advice certified 30 completions versus 9, a 36.2‑percentage‑point difference (exact family sign‑flip p=0.0078), using 272 proposals versus 399. All 116 trajectories passed independent replay. Both models showed a positive completion difference. The comparison measures the evaluated advice interface under a common shield; search‑versus‑rule advice requires a separate matched comparison. Unknown and handoff outcomes remain explicit, with guarantees tied to declared provider capabilities rather than arbitrary third‑party exactly‑once execution.\n\nReview: The study demonstrates that leveraging public evidence for task recovery under uncertain tool behavior is feasible and yields a clear advantage over conventional prompting methods.