Policies

Home > The Dockfile > Policies

Policies give you control over what your agent can do and what it outputs. The policies section in the Dockfile configures tool access restrictions and safety guardrails. Policies are enforced by the RuntimePolicyEngine in the runtime.

Overview

policies:
  tools:
    allowed: [web_search, calculator]
    deny_by_default: true
  safety:
    block_prompt_injection: true
    redact_patterns:
      - '\b\d{3}-\d{2}-\d{4}\b'    # SSN
      - '\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b'  # email
    max_output_chars: 10000
    halt_on_violation: false

Both tools and safety are optional. If the policies section is omitted, no tool restrictions or safety filters are applied (except block_prompt_injection defaults to true when safety is present).

How Policies Wire into the Runtime

1. The RuntimePolicyEngine is created from spec.policies during app startup
2. Before invocation: validate_input() checks for prompt injection patterns
3. After invocation: apply_output_policies() runs redaction, then truncation
4. Tool calls: the adapter can check is_tool_allowed() before executing tools

If halt_on_violation is true, a policy violation raises an error instead of silently filtering.

In This Section

Page	What it covers
Prompt Injection Blocking	Built-in patterns, block_prompt_injection flag
Output Controls	redact_patterns (regex-based) and max_output_chars
Tool Gating	tools.allowed, deny_by_default

Source: Policies, ToolPolicy, SafetyPolicy in packages/schema/dockrion_schema/dockfile_v1.py; RuntimePolicyEngine in packages/runtime/dockrion_runtime/policies.py

---

Up: The Dockfile | Previous: 2.3 Auth | Next: 2.5 Secrets →