Skip to main content

AGT Policy YAML Reference

Status

Owner: TBD · Status: draft · Canonical source: flezi.app.ai.agent/docs/governance/agt-policy-overview.md

Comprehensive specification of the Agent Governance Toolkit (AGT) Policy YAML format used to govern agent tool calls.

General Syntax​

version: "1.0"
name: "policy_name"
description: "Human-readable description"
rules:
  - tool: "tool_name_or_wildcard"
    action: "allow | deny | require_approval"
    reason: "Reason displayed when rule triggers"
    parameters: # Optional dynamic parameter checks
      parameter_name:
        matches: "regex_pattern"

Available Actions​

ActionMeaningBehavioral Outcome
allowPermit executionTool executes immediately without friction
denyProhibit executionIntercepted immediately; error reported to model
require_approvalMandatory human sign-offExecution pauses; renders an approval banner in chat

Sample Policy​

version: "1.0"
name: "customer_facing_safety_policy"
rules:
  # 1. Block shell command execution
  - tool: "bash"
    action: deny
    reason: "Customer-facing agents must not execute OS shell commands"

  # 2. Allow knowledge base queries
  - tool: "query_knowledge_base"
    action: allow

  # 3. Require approval for database modifications
  - tool: "sql_execute"
    action: require_approval
    parameters:
      query:
        matches: "(?i)(DROP|DELETE|UPDATE|ALTER)"
    reason: "Database mutation requires human review"

  # 4. Allow all remaining tools by default
  - tool: "*"
    action: allow