发现最适合你需求的 AI 技能
Normalize completion claims, expected postconditions and evidence references into a reviewable local contract.
Detect partial completion, silent no-op, stale evidence and contradictions between claims, tool results and read-back facts.
Build a human-review receipt for verified, partial or contradicted completion claims without releasing external changes.
Compare supplied authoritative read-back observations with expected postconditions without querying a live system.
Map detected loop patterns to cache, rewrite, block or human-escalation decisions.
Normalize redacted agent tool events into comparable call, result and state fingerprints.
Detect repeated calls, retry storms, cycles and no-progress states in a normalized agent trace.
Build a pending replan and human-review packet without replaying side effects.
Detect ambiguous and nonexistent local times at daylight-saving transitions before scheduling.
Build a canonical UTC instant and recipient-local renderings without writing to a calendar.
Hold a timezone-sensitive schedule for human review with explicit evidence and no external side effects.
Normalize an authorized local-time request into an explicit IANA timezone contract.