Wave-0 demo — agentpolicy -> agentinbox -> human deny
agentinbox 0.1.0.dev0; agentpolicy 0.1.0
1) Replit destructive rule: drop_database -> deny (deny_critical_op)
   drop_database (delete) in production on restricted data is blocked: destructive operation on sensitive data with irreversible side effects. Use a read-only alternative or move to a lower-risk environment. Policy rule: Deletes in production are blocked.
2) Production-write human gate: deploy_service -> escalate (escalate_prod_write)
   Write action (deploy) in production on restricted data requires approval. Route to the on-call approver. Policy rule: Write actions in production require approval.
3) ask-raised: ask_ef87eb1d5c4b896a (session sess_wave0_demo)
ASKS: ask_ef87eb1d5c4b896a   status=pending   ttl=4m59s
fingerprint: sha256:b4c68e7261ce…

CHOICES
  suggestion: run against staging first (suggestion)
  suggestion: cap the blast radius with an explicit limit (suggestion)

ACTION
  tool: [untrusted: tool output] shell
  action: [untrusted: tool output] run_command
  arguments: [untrusted: tool output] command=deploy_service checkout-api; timeout_s=30
  fingerprint: sha256:b4c68e7261ce…
  class: irreversible
  criticality: high
  requested by: release-bot

POLICY
  reason code: human_gate_required
  rule: rule[2]
  policy version: 1.0.0
  clause: [untrusted: tool output] rule[2]

BLAST RADIUS
  resource: service:checkout-api
  reversibility: irreversible
  affected estimate: unknown
  environment: production
  data class: restricted
  credential class: unknown

SESSION SO FAR
  session: sess_wave0_demo
  agent: agent://release-bot
  tier: unknown
  escalations: none
  prior asks: none
  similar cases: none (precedent only)
  prior objection: none
  session tail: unknown
  flags: code_freeze_window
  alternatives: staging deploy, read-only status check
  transitions: 1

Answer with: agentinbox answer <id> --approve|--deny  (q leaves it pending)
4) delivered: terminal card + desktop notification
ASKS: ask_ef87eb1d5c4b896a   status=pending   ttl=4m59s
fingerprint: sha256:b4c68e7261ce…

CHOICES
  suggestion: run against staging first (suggestion)
  suggestion: cap the blast radius with an explicit limit (suggestion)

ACTION
  tool: [untrusted: tool output] shell
  action: [untrusted: tool output] run_command
  arguments: [untrusted: tool output] command=deploy_service checkout-api; timeout_s=30
  fingerprint: sha256:b4c68e7261ce…
  class: irreversible
  criticality: high
  requested by: release-bot

POLICY
  reason code: human_gate_required
  rule: rule[2]
  policy version: 1.0.0
  clause: [untrusted: tool output] rule[2]

BLAST RADIUS
  resource: service:checkout-api
  reversibility: irreversible
  affected estimate: unknown
  environment: production
  data class: restricted
  credential class: unknown

SESSION SO FAR
  session: sess_wave0_demo
  agent: agent://release-bot
  tier: unknown
  escalations: none
  prior asks: none
  similar cases: none (precedent only)
  prior objection: none
  session tail: unknown
  flags: code_freeze_window
  alternatives: staging deploy, read-only status check
  transitions: 1

Answer with: agentinbox answer <id> --approve|--deny  (q leaves it pending)
ask ask_ef87eb1d5c4b896a: denied
5) terminal decision: deny (exit 2)
6) resolve: resolve/1.0 deny (deny_ask_denied)
7) agentwatch events: ['ask-raised', 'ask-delivered', 'ask-delivered', 'ask-answered']
8) export: 10 records
9) verify-log: ok: 5 audit entries verify (exit 0)
