Policies

Policies are the core of Sequenceware. They define rules that match tool calls by any combination of fields and enforce actions automatically.

How policies work

When an agent makes a tool call, Sequenceware evaluates it against all enabled policies:

Tool call received → Match against all policies → Apply strictest action

Strictest-action-wins: when multiple policies match, the most restrictive action applies.

block > require_approval > warn > allow

Policy actions

Action	Behavior	Use case
allow	Tool call proceeds normally	Explicitly allow known-safe operations
warn	Tool call proceeds, risk flagged in dashboard	Monitor without blocking
block	Tool call is immediately rejected	Prevent dangerous operations
require_approval	Tool call pauses until a human approves	Critical operations needing review

Match conditions

Each policy has one or more match conditions. All conditions must match for the policy to trigger (AND logic).

Fields

You can match on any field in the tool call event:

Field	Description	Example value
toolName	Name of the tool being called	create_pull_request
agentId	ID of the agent	claude-code-maria
inputPayload.*	Any field in the tool's input	inputPayload.environment
metadata.*	Any metadata field	metadata.branch
actionHint	Semantic action hint	deploy, pr_merge

Operators

Operator	Description	Example
equals	Exact match	toolName equals "deploy"
not_equals	Not equal	agentId not_equals "admin-agent"
contains	Substring match	toolName contains "delete"
matches	Regex match	inputPayload.query matches "(DROP|TRUNCATE)"
in	Value in list	inputPayload.environment in ["production", "prod"]
not_in	Value not in list	agentId not_in ["trusted-agent"]
exists	Field exists	inputPayload.password exists
gt / lt / gte / lte	Numeric comparison	inputPayload.amount gt 1000

Default policies

Sequenceware ships with sensible defaults that are created on first boot:

Policy	Matches	Action
Sensitive Files Gate	.env, credentials, secrets, .pem	require_approval
Deployment Gate	Deploy actions to production	require_approval
Destructive Script Block	rm -rf, DROP TABLE, destructive commands	block
Main Branch Protection	Direct pushes/merges to main/master	require_approval

Creating policies

Via the dashboard

1. Navigate to Policies > Create Policy
2. Add match conditions
3. Select the action
4. Enable the policy

Via the API

curl -X POST http://localhost:3000/v1/policies \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Block prod database writes",
    "description": "Prevent agents from writing to production databases",
    "matchConditions": [
      { "field": "toolName", "operator": "equals", "value": "db_query" },
      { "field": "inputPayload.environment", "operator": "equals", "value": "production" },
      { "field": "inputPayload.operation", "operator": "in", "value": ["INSERT", "UPDATE", "DELETE"] }
    ],
    "action": "block",
    "enabled": true
  }'

Via the SDK

await client.createPolicy({
  name: 'Block prod database writes',
  description: 'Prevent agents from writing to production databases',
  matchConditions: [
    { field: 'toolName', operator: 'equals', value: 'db_query' },
    { field: 'inputPayload.environment', operator: 'equals', value: 'production' },
  ],
  action: 'block',
  enabled: true,
});

Policy simulation

Test policies before enabling them using the policy simulator:

curl -X POST http://localhost:3000/v1/policies/simulate \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "toolName": "deploy",
    "inputPayload": {
      "environment": "production",
      "service": "api"
    }
  }'

The simulator returns which policies would match and what action would be taken — without affecting any real agent.

Rate limiting

Policies can include rate limits to control how frequently agents can use specific tools:

{
  "name": "Rate limit API writes",
  "matchConditions": [
    { "field": "toolName", "operator": "matches", "value": "(create_|update_|delete_)" }
  ],
  "action": "block",
  "rateLimit": {
    "maxCalls": 50,
    "windowMinutes": 60,
    "scope": "per-agent"
  }
}

Rate limit scopes:

• per-agent — limit applies to each agent individually
• per-tool — limit applies to each tool across all agents
• per-org — limit applies organization-wide

Policy versioning

Policies are automatically versioned. Every update increments the version number and preserves the previous state. This gives you a full history of how a policy evolved over time.

Each policy includes:

• version — auto-incremented integer starting at 1
• updatedAt — timestamp of the last modification

View the full version history of any policy:

curl http://localhost:3000/v1/policies/pol_abc123/versions \
  -H "Authorization: Bearer $TOKEN"

This is useful for:

• Auditing — understanding why a policy changed and who changed it
• Debugging — identifying when a policy change caused unexpected behavior
• Compliance — maintaining a record of governance rule evolution

Policy presets

Sequenceware includes preset policies grouped by category that you can enable with one click from the dashboard. View available presets at GET /v1/policy-presets.