Appendix A — Agent Records and Handoff Schemas
Copy-paste record shapes for handoffs, decision events, and role definitions used throughout the book.
Handoff record
One record per transfer between roles. Every claim carries its evidence status and source.
{
"role": "researcher",
"task_id": "urban-transport-001",
"claim": "Micromobility is increasingly regulated through city-level safety rules.",
"evidence": "source URL or document path",
"evidence_type": "background | source | observation | derivation",
"status": "unverified | verifying | verified | disputed | superseded",
"confidence": "low | medium | high",
"open_questions": ["Which cities changed policy in the last 12 months?"],
"requested_action": "verify source date",
"provenance": "retrieval query + document id, or step that produced it"
}
Decision event
The minimum trace that lets you answer later: who decided, from which state, using what evidence, and what changed?
decision_id
run_id
actor
input_state_id
proposed_action
authority_check (allowed | needs review | rejected, with policy reference)
evidence_used
accepted_or_rejected
output_state_id
Role record
name: critic
version: 1.3.0
model: recorded_at_runtime
prompt_id: critic_prompt_v4
input_schema: candidate_summary_v2
output_schema: critique_v1
tools: [citation_checker]
memory_access: [current_task_log]
authority: recommend_only
status: production
Memory-entry fields
content
stored_by / stored_at
source / provenance
valid_at / known_at (when true / when recorded)
supersedes / superseded_by
scope (which project, task, or context may use it)
expiry or review date
These shapes are starting points. Their value is not the field list — it is that every handoff, decision, and memory becomes inspectable by someone other than its author.