Skip to main content
The policy engine lets you define rules that your audit trail must satisfy, then check them automatically.

​
Built-in policies

Five policies are included and run with a single command: 
plog policy check
PolicyDescription
Chain IntegrityHash chain is unbroken
Event CompletenessEvery TOOL_CALL has a matching TOOL_RESULT
Session HealthNo >1 hour gaps between consecutive events
Error RateFlag sessions with >50% error events
Duration OutliersFlag tool calls >60 seconds

​
Custom policies (YAML)

Define custom policies in a YAML file: 
plog policy check --file policies.yaml
plog policy check --file policies.yaml --policy my-policy  # Run one specific policy

​
Rule types

The policy engine supports 7 rule types:

​
Sequence

If event X occurs, event Y must follow within a duration: 
policies:
  - name: tool-result-follows-call
    rules:
      - name: result-after-call
        type: sequence
        when:
          action_type: TOOL_CALL
        expect:
          action_type: TOOL_RESULT
        within: "60s"

​
Absence

If event X occurs, event Y must NOT follow: 
policies:
  - name: no-delete-after-read
    rules:
      - name: no-delete
        type: absence
        when:
          action_name: read_file
        deny:
          action_name: delete_file

​
Boundary

Restrict an agent to a set of allowed actions: 
policies:
  - name: agent-boundary
    rules:
      - name: restricted-actions
        type: boundary
        allow_actions:
          - read_file
          - search

​
Presence

A scope must contain at least one matching event: 
policies:
  - name: must-have-llm-call
    rules:
      - name: require-llm
        type: presence
        scope: session
        require:
          action_type: LLM_CALL

​
Rate

No more than X events of type Y per scope: 
policies:
  - name: rate-limit-api-calls
    rules:
      - name: llm-rate
        type: rate
        per: session
        count:
          action_type: LLM_CALL
        max: 100

​
Pattern

Field values must match (or not match) a regex: 
policies:
  - name: no-pii-in-output
    rules:
      - name: no-ssn
        type: pattern
        for_field: action_output
        deny_pattern: "\\d{3}-\\d{2}-\\d{4}"
Use match_pattern to require a field value matches a regex, or deny_pattern to require it does not.

​
Temporal

Events must fall within a time window: 
policies:
  - name: business-hours-only
    rules:
      - name: weekday-hours
        type: temporal
        when:
          action_type: TOOL_CALL
        hours: [9, 17]
        days: [1, 2, 3, 4, 5]
        timezone: "America/New_York"
The hours field takes [start, end) in 24-hour format. The days field uses 0=Sunday through 6=Saturday.

​
Common rule fields

All rule types support these optional fields:
FieldDescription
nameName of the rule
severity"fail" (default) or "warn"
scope"session" or "agent"
label_filterFilter events by label values before evaluation

​
Dry run

Test policies against existing events without affecting anything: 
plog policy test --file policies.yaml
plog policy test requires --file — it is designed for validating custom policy definitions.

​
Policy result schema

{
  "policy": "event-completeness",
  "status": "fail",
  "violations": [
    {
      "policy_name": "event-completeness",
      "rule": "result-after-call",
      "trigger_event": "evt-123",
      "expected_event": "TOOL_RESULT",
      "session_id": "session-456",
      "agent_id": "my-agent",
      "timestamp": "2026-02-17T10:30:00Z",
      "message": "TOOL_CALL without matching TOOL_RESULT",
      "detail": "Expected TOOL_RESULT within 60s of TOOL_CALL at sequence 42"
    }
  ],
  "events_evaluated": 150,
  "timestamp": "2026-02-17T12:00:00Z"
}