Policy Schema Reference - Agent Identity Protocol

Complete reference for AIP policy YAML schema with all fields, types, defaults, and validation rules. This page covers both v1alpha1 and v1alpha2.

For a user-friendly guide to writing policies, see the Policy Reference. For formal specifications, see v1alpha1 or v1alpha2.

Document Structure

v1alpha2 (Current)
v1alpha1

apiVersion: aip.io/v1alpha2
kind: AgentPolicy
metadata:
  name: <string>
  version: <string>
  owner: <string>
  signature: <string>
spec:
  mode: <string>
  allowed_tools: [<string>]
  allowed_methods: [<string>]
  denied_methods: [<string>]
  tool_rules: [<ToolRule>]
  protected_paths: [<string>]
  strict_args_default: <bool>
  dlp: <DLPConfig>
  identity: <IdentityConfig>
  server: <ServerConfig>

apiVersion: aip.io/v1alpha1
kind: AgentPolicy
metadata:
  name: <string>
  version: <string>
  owner: <string>
spec:
  mode: <string>
  allowed_tools: [<string>]
  allowed_methods: [<string>]
  denied_methods: [<string>]
  tool_rules: [<ToolRule>]
  protected_paths: [<string>]
  strict_args_default: <bool>
  dlp: <DLPConfig>

Top-Level Fields

apiVersion

string

required

API version of the policy schema.Valid values:

aip.io/v1alpha1 - Original specification
aip.io/v1alpha2 - Current specification with identity and server features

Validation: Implementations MUST reject documents with unknown apiVersion.

Example:

apiVersion: aip.io/v1alpha2

kind

string

required

Resource kind. MUST be AgentPolicy.

Example:

kind: AgentPolicy

metadata

Policy metadata for identification and integrity verification.

metadata.name

string

required

Unique identifier for this policy.Format: DNS-1123 subdomain (lowercase alphanumeric + hyphens)

Must start and end with alphanumeric
Max length: 253 characters
Pattern: ^[a-z0-9]([-a-z0-9]*[a-z0-9])?$

Examples:

metadata:
  name: production-agent
  # OR
  name: dev-env-policy
  # OR
  name: team-123-policy

metadata.version

string

Semantic version of the policy.Format: MAJOR.MINOR.PATCH[-PRERELEASE]

Pattern: ^[0-9]+\.[0-9]+\.[0-9]+(-[a-zA-Z0-9]+)?$

Use case: Track policy changes, audit compliance.

Examples:

version: "1.0.0"
# OR
version: "2.1.3-beta"

metadata.owner

string

Contact email for policy questions.Format: Valid email address

Example:

owner: security-team@company.com

metadata.signature (v1alpha2)

metadata.signature

string

Cryptographic signature for policy integrity verification.Format: <algorithm>:<base64-encoded-signature>Supported algorithms:

ed25519 - Ed25519 signature (RECOMMENDED)
ecdsa-p256 - ECDSA with P-256 curve

Validation: When present, implementations MUST verify the signature before applying the policy.

Example:

signature: "ed25519:YWJjZGVmZ2hpamtsbW5vcHFyc3R1dnd4eXoxMjM0NTY3ODkwYWJjZGVm"

spec

Policy specification defining authorization rules.

spec.mode

enum

default:"enforce"

Enforcement mode for policy violations.Values:

enforce - Block violations, return JSON-RPC error (default)
monitor - Log violations but allow through (dry-run)

Use case: Use monitor to test new policies before enforcement.

Example:

spec:
  mode: enforce

spec.allowed_tools

array

Allowlist of tool names that the agent MAY invoke.Item type: String (tool name)Validation:

Must be unique
Minimum length: 1
Subject to name normalization (NFKC, lowercase)

Default deny: Tools not in this list AND not in tool_rules with action: allow are blocked.

Example:

allowed_tools:
  - github_get_repo
  - github_list_pulls
  - read_file
  - list_directory

spec.allowed_methods

array

Allowlist of JSON-RPC methods that are permitted.Item type: String (method name)Default: If not specified, implementations use a safe default list (see below).Wildcard: * allows all methods.

Default allowed methods:

allowed_methods:
  - initialize
  - initialized
  - ping
  - tools/call
  - tools/list
  - completion/complete
  - notifications/initialized
  - notifications/progress
  - notifications/message
  - notifications/resources/updated
  - notifications/resources/list_changed
  - notifications/tools/list_changed
  - notifications/prompts/list_changed
  - cancelled

spec.denied_methods

array

Denylist of JSON-RPC methods that are explicitly denied.Item type: String (method name)Precedence: Denied methods take precedence over allowed methods.

Example:

denied_methods:
  - resources/read
  - resources/write
  - prompts/get

spec.protected_paths

array

File paths that tools MUST NOT access.Item type: String (file path)Behavior:

Any tool argument containing a protected path is BLOCKED
Paths are expanded (~ → home directory)
Policy file itself is automatically protected

Example:

protected_paths:
  - ~/.ssh
  - ~/.aws/credentials
  - .env
  - /etc/passwd

spec.strict_args_default

boolean

default:"false"

When true, tool rules reject any arguments not explicitly declared in allow_args.Use case: Enforce strict argument validation for all tools.

Example:

strict_args_default: true

spec.tool_rules

Fine-grained rules for individual tools.

Tool Rule Structure

Complete Tool Rule Schema

tool_rules:
  - tool: <string>              # REQUIRED - Tool name
    action: <string>            # OPTIONAL - allow|block|ask (default: allow)
    rate_limit: <string>        # OPTIONAL - e.g., "10/minute"
    strict_args: <bool>         # OPTIONAL - Override strict_args_default
    schema_hash: <string>       # OPTIONAL - Tool schema integrity (v1alpha2)
    allow_args:                 # OPTIONAL - Argument validation
      <arg_name>: <regex>

tool_rules[].tool

string

required

Tool name this rule applies to.Validation: Subject to name normalization (NFKC, lowercase).

Example:

tool_rules:
  - tool: read_file

tool_rules[].action

enum

default:"allow"

Action to take when this tool is invoked.Values:

allow - Permit (subject to allow_args validation)
block - Deny unconditionally
ask - Require interactive user approval

Examples:

# Block destructive operations
- tool: delete_file
  action: block

# Require approval for sensitive operations
- tool: run_training
  action: ask

# Allow with argument validation
- tool: exec_command
  action: allow
  allow_args:
    command: "^(ls|cat|echo|pwd)\s.*"

tool_rules[].rate_limit

string

Rate limiting for this tool.Format: <count>/<period>Periods:

second (aliases: sec, s)
minute (aliases: min, m)
hour (aliases: hr, h)

Algorithm: Implementation-defined (token bucket, sliding window, etc.)

Examples:

- tool: list_gpus
  rate_limit: "10/minute"

- tool: search_files
  rate_limit: "100/hour"

- tool: api_request
  rate_limit: "5/second"

tool_rules[].strict_args

boolean

Override spec.strict_args_default for this specific tool.When true, reject any arguments not declared in allow_args.

Example:

- tool: postgres_query
  strict_args: true
  allow_args:
    query: "^SELECT\s+.*"  # Only SELECT queries
    # Any other argument will be rejected

tool_rules[].schema_hash (v1alpha2)

tool_rules[].schema_hash

string

Cryptographic hash of the tool’s MCP schema for integrity verification.Format: <algorithm>:<hex-digest>Algorithms:

sha256 (RECOMMENDED)
sha384
sha512

Use case: Prevent tool poisoning attacks by verifying the tool definition hasn’t changed.

Example:

- tool: read_file
  schema_hash: "sha256:a3c7f2e8d9b4f1e2c8a7d6f3e9b2c4f1a8e7d3c2b5f4e9a7c3d8f2b6e1a9c4f7"

Generating schema hashes:

# Using AIP CLI
aip-proxy schema-hash --server mcp://localhost:8080 --tool read_file

tool_rules[].allow_args

object

Argument validation using regex patterns.Format: Map of argument name → regex patternBehavior:

All arguments in this map MUST be present in the request
Argument values MUST match their regex pattern
Missing or non-matching arguments result in BLOCK

Regex engine: MUST support RE2 semantics (linear-time guarantees)

Examples:

allow_args:
  # Constrain URLs to GitHub only
  url: "^https://github\.com/.*"
  
  # Allow only SELECT queries
  query: "^SELECT\s+.*"
  
  # Restrict file paths to home directory
  path: "^/home/.*"
  
  # Validate email format
  email: "^[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}$"

String conversion: Non-string argument values are converted before matching:

String → as-is
Number → decimal representation
Boolean → “true” or “false”
Null → empty string
Array/Object → JSON serialization

spec.dlp

Data Loss Prevention configuration for scanning and redacting sensitive data.

Complete DLP Schema

v1alpha2
v1alpha1

dlp:
  enabled: <bool>             # default: true when dlp block present
  scan_requests: <bool>       # default: false
  scan_responses: <bool>      # default: true
  detect_encoding: <bool>     # default: false
  filter_stderr: <bool>       # default: false
  max_scan_size: <string>     # default: "1MB"
  on_request_match: <string>  # default: "block"
  patterns:
    - name: <string>
      regex: <string>
      scope: <string>         # default: "all"

dlp:
  enabled: <bool>             # default: true when dlp block present
  detect_encoding: <bool>     # default: false
  filter_stderr: <bool>       # default: false
  patterns:
    - name: <string>
      regex: <string>

dlp.enabled

boolean

default:"true"

Enable DLP scanning.Default: true when dlp block is present

dlp.scan_requests (v1alpha2)

dlp.scan_requests

boolean

default:"false"

Scan tool arguments for sensitive data before forwarding to MCP server.Use case: Prevent data exfiltration via arguments (e.g., embedding secrets in API queries).

dlp.scan_responses

boolean

default:"true"

Scan tool responses for sensitive data.Use case: Standard DLP to prevent PII/secrets from reaching the agent.

dlp.max_scan_size (v1alpha2)

dlp.max_scan_size

string

default:"1MB"

Maximum size of content to scan per request/response.Format: Size string (e.g., "1MB", "512KB", "10MB")Purpose: Prevents ReDoS and memory exhaustion on large payloads.Behavior: Content exceeding this limit is truncated for scanning, and a warning is logged.

dlp.on_request_match (v1alpha2)

dlp.on_request_match

enum

default:"block"

Action when DLP pattern matches in a request (requires scan_requests: true).Values:

block - Reject the request with error -32001 (default)
redact - Replace matched content and forward
warn - Log warning and forward unchanged

dlp.patterns

array

required

List of DLP detection patterns.Item type: Pattern object with name and regex

dlp.patterns[].name

string

required

Human-readable identifier for this pattern.Use case: Appears in audit logs and redaction markers.

dlp.patterns[].regex

string

required

Regular expression to detect sensitive data.Regex engine: MUST support RE2 semantics (linear-time guarantees)

dlp.patterns[].scope (v1alpha2)

dlp.patterns[].scope

enum

default:"all"

Where to apply this pattern.Values:

all - Scan both requests and responses (default)
request - Only scan requests (detect exfiltration attempts)
response - Only scan responses (PII protection)

Example patterns:

dlp:
  scan_requests: true
  patterns:
    # API Keys
    - name: "AWS Key"
      regex: "(A3T[A-Z0-9]|AKIA|AGPA|AIDA|AROA|AIPA|ANPA|ANVA|ASIA)[A-Z0-9]{16}"
      scope: "all"
    
    - name: "GitHub Token"
      regex: "ghp_[a-zA-Z0-9]{36}"
      scope: "all"
    
    # PII (response-only)
    - name: "Email"
      regex: "[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}"
      scope: "response"
    
    - name: "SSN"
      regex: "\b\d{3}-\d{2}-\d{4}\b"
      scope: "response"
    
    # Injection detection (request-only)
    - name: "SQL Injection"
      regex: "(?i)(DROP|DELETE|TRUNCATE)\s+TABLE"
      scope: "request"

Redaction format: Matched content is replaced with:

[REDACTED:<name>]

Example:

Before: "Connect with: AKIAIOSFODNN7EXAMPLE"
After:  "Connect with: [REDACTED:AWS Key]"

spec.identity (v1alpha2)

Agent identity and session management configuration.

Complete Identity Schema

identity:
  enabled: <bool>           # default: false
  token_ttl: <duration>     # default: "5m"
  rotation_interval: <duration>  # default: "4m"
  require_token: <bool>     # default: false
  session_binding: <string> # default: "process"
  nonce_window: <duration>  # default: equals token_ttl
  policy_transition_grace: <duration>  # default: "0s"
  audience: <string>        # default: metadata.name
  nonce_storage:            # optional
    type: <string>          # default: "memory"
    address: <string>
    key_prefix: <string>    # default: "aip:nonce:"
    clock_skew_tolerance: <duration>  # default: "30s"
  keys:                     # optional
    signing_algorithm: <string>  # default: "ES256"
    key_source: <string>    # default: "generate"
    key_path: <string>
    rotation_period: <duration>  # default: "7d"

identity.enabled

boolean

default:"false"

Enable identity token generation and management.

identity.token_ttl

duration

default:"5m"

Token time-to-live.Format: Go duration string (e.g., "5m", "1h", "300s")Recommendation: Use short TTLs (5-15 minutes) to limit token theft window.

identity.rotation_interval

duration

default:"4m"

How often to rotate tokens before expiry.Constraint: MUST be less than token_ttlRecommendation: Leave 1-2 minute grace period for in-flight requests.

identity.require_token

boolean

default:"false"

Require valid identity token for all tool calls.Error code: Requests without tokens are rejected with -32008.Use case: Gradual rollout - start with false to generate tokens without enforcement.

identity.session_binding

enum

default:"process"

Context bound to session identity.Values:

process - Bind to OS process ID (default)
policy - Bind to policy hash
strict - Bind to process + policy + hostname

Use case:

process: Single-machine, local agents
policy: Distributed agents sharing policy
strict: High security, non-ephemeral environments

identity.audience

string

Intended audience for identity tokens (JWT aud claim).Default: Value of metadata.nameFormat: URI string identifying the MCP server or servicePurpose: Prevents tokens issued for one MCP server from being accepted by another.

Example:

identity:
  enabled: true
  audience: "https://mcp.example.com/api"

identity.nonce_storage

Distributed nonce storage configuration for multi-instance deployments.

identity.nonce_storage.type

enum

default:"memory"

Storage backend for nonces.Values:

memory - In-memory sync.Map (single-instance only)
redis - Redis cluster (RECOMMENDED for production)
postgres - PostgreSQL database

identity.nonce_storage.address

string

Connection string for external storage.Required: When type is redis or postgres

Examples:

# Redis cluster
identity:
  nonce_storage:
    type: "redis"
    address: "redis://redis-cluster:6379"
    key_prefix: "prod:aip:nonce:"

# PostgreSQL
identity:
  nonce_storage:
    type: "postgres"
    address: "postgres://user:pass@db:5432/aip?sslmode=require"

Using type: "memory" with multiple AIP instances is a security vulnerability that allows cross-instance replay attacks.

spec.server (v1alpha2)

HTTP server for remote validation endpoints.

Complete Server Schema

server:
  enabled: <bool>           # default: false
  listen: <string>          # default: "127.0.0.1:9443"
  failover_mode: <string>   # default: "fail_closed"
  timeout: <duration>       # default: "5s"
  tls:
    cert: <string>
    key: <string>
  endpoints:
    validate: <string>      # default: "/v1/validate"
    revoke: <string>        # default: "/v1/revoke"
    jwks: <string>          # default: "/v1/jwks"
    health: <string>        # default: "/health"
    metrics: <string>       # default: "/metrics"

server.enabled

boolean

default:"false"

Enable HTTP server for remote validation.

server.listen

string

default:"127.0.0.1:9443"

Address and port to bind the HTTP server.Format: <host>:<port> or :<port>

Binding to 0.0.0.0 exposes the validation endpoint to the network. Implementations MUST require TLS when listen address is not localhost.

server.failover_mode

enum

default:"fail_closed"

Behavior when validation server is unreachable.Values:

fail_closed - Deny all requests (RECOMMENDED for production)
fail_open - Allow all requests (NOT RECOMMENDED)
local_policy - Fall back to local policy evaluation (RECOMMENDED for hybrid)

server.timeout

duration

default:"5s"

Maximum time to wait for validation server response.After timeout, failover_mode behavior is triggered.

server.tls

TLS configuration for HTTPS.

server.tls.cert

string

Path to TLS certificate (PEM format).Required: When listen is not localhost

server.tls.key

string

Path to TLS private key (PEM format).Required: When listen is not localhost

Example:

server:
  enabled: true
  listen: "0.0.0.0:9443"
  tls:
    cert: "/etc/aip/tls/cert.pem"
    key: "/etc/aip/tls/key.pem"

JSON Schema Validation

Validate your policy file against the JSON Schema:

# Using ajv-cli
npm install -g ajv-cli
ajv validate -s spec/schema/agent-policy-v1alpha2.schema.json -d policy.yaml

# Using Python jsonschema
pip install jsonschema pyyaml
python -c "
import yaml, jsonschema, json
schema = json.load(open('spec/schema/agent-policy-v1alpha2.schema.json'))
policy = yaml.safe_load(open('policy.yaml'))
jsonschema.validate(policy, schema)
print('Valid!')
"

Common Validation Errors

Error	Cause	Fix
`invalid apiVersion`	Wrong API version	Use `aip.io/v1alpha1` or `aip.io/v1alpha2`
`empty allowed_tools`	No tools specified	Add tools or tool_rules
`invalid regex in allow_args`	Bad regex pattern	Validate regex syntax (RE2)
`invalid rate_limit format`	Wrong rate limit format	Use `<N>/<period>`
`rotation_interval >= token_ttl`	Rotation must happen before expiry	Reduce rotation_interval
`missing metadata.name`	Required field missing	Add unique policy name

Specifications

API Reference

SDKs & Implementations

Documentation Index

​Document Structure

​Top-Level Fields

​apiVersion

​kind

​metadata

​metadata.name

​metadata.version

​metadata.owner

​metadata.signature (v1alpha2)

​spec

​spec.mode

​spec.allowed_tools

​spec.allowed_methods

​spec.denied_methods

​spec.protected_paths

​spec.strict_args_default

​spec.tool_rules

​Tool Rule Structure

​tool_rules[].tool

​tool_rules[].action

​tool_rules[].rate_limit

​tool_rules[].strict_args

​tool_rules[].schema_hash (v1alpha2)

​tool_rules[].allow_args

​spec.dlp

​dlp.enabled

​dlp.scan_requests (v1alpha2)

​dlp.scan_responses

​dlp.max_scan_size (v1alpha2)

​dlp.on_request_match (v1alpha2)

​dlp.patterns

​dlp.patterns[].name

​dlp.patterns[].regex

​dlp.patterns[].scope (v1alpha2)

​spec.identity (v1alpha2)

​identity.enabled

​identity.token_ttl

​identity.rotation_interval

​identity.require_token

​identity.session_binding

​identity.audience

​identity.nonce_storage

​spec.server (v1alpha2)

​server.enabled

​server.listen

​server.failover_mode

​server.timeout

​server.tls

​JSON Schema Validation

​Common Validation Errors

​See Also

Build docs developers (and LLMs) love

Document Structure

Top-Level Fields

apiVersion

kind

metadata

metadata.name

metadata.version

metadata.owner

metadata.signature (v1alpha2)

spec

spec.mode

spec.allowed_tools

spec.allowed_methods

spec.denied_methods

spec.protected_paths

spec.strict_args_default

spec.tool_rules

Tool Rule Structure

tool_rules[].tool

tool_rules[].action

tool_rules[].rate_limit

tool_rules[].strict_args

tool_rules[].schema_hash (v1alpha2)

tool_rules[].allow_args

spec.dlp

dlp.enabled

dlp.scan_requests (v1alpha2)

dlp.scan_responses

dlp.max_scan_size (v1alpha2)

dlp.on_request_match (v1alpha2)

dlp.patterns

dlp.patterns[].name

dlp.patterns[].regex

dlp.patterns[].scope (v1alpha2)

spec.identity (v1alpha2)

identity.enabled

identity.token_ttl

identity.rotation_interval

identity.require_token

identity.session_binding

identity.audience

identity.nonce_storage

spec.server (v1alpha2)

server.enabled

server.listen

server.failover_mode

server.timeout

server.tls

JSON Schema Validation

Common Validation Errors

See Also