Agent Tool Policy Filter

Overview

Applies only the allowlists, denylists, tag requirements, scope allowance and cost, latency or side-effect ceilings you provide. Surviving tools keep input order. The skill reads no task text, computes no relevance score, recommends nothing, does not consult Squillify entitlements, and does not authorise execution. Every reported exclusion names the exact policy rule that caused it.

Prompt assembly often needs a small, auditable policy gate before tools are shown to a model. The mechanical filter makes that step reproducible without pretending to understand which tool is relevant.

When to use it

Scope

Non-goals

• Does not rank, score, route or recommend tools
• Does not read task text or infer caller intent
• Does not consult the live catalog, tenant entitlements, quota or authorization
• Does not execute any tool

Properties

Capabilities

Cost & billing

Flat rate per invocation

File handling

How we handle your data

Input schema

Output schema

API contract

curl -sS -X POST "$BASE/v1/execute" -H "Authorization: Bearer $KEY" -H 'Content-Type: application/json' -d '{"skillKey":"agent.tool_policy_filter","input":{"toolMetadata":[{"key":"text.clean","sideEffect":"NONE"}],"constraints":{"maxSideEffect":"NONE"}}}'

import { Squillify } from "@squillify/sdk";

const sq = new Squillify({ apiKey: process.env.SQUILLIFY_API_KEY });

const { output } = await sq.execute("agent.tool_policy_filter", {"toolMetadata":[{"key":"mail.send_summary","requiredScopes":["mail:send"],"sideEffect":"EXTERNAL_WRITE"},{"key":"text.build_summary","requiredScopes":["skills:invoke"],"sideEffect":"NONE"}],"constraints":{"allowedScopes":["skills:invoke"],"maxSideEffect":"NONE"},"includeExcluded":true});

Errors

Other limits

Limitations

• Does not rank, score, route or recommend tools
• Does not read task text or infer caller intent
• Does not consult the live catalog, tenant entitlements, quota or authorization
• Does not execute any tool
• Choosing the most relevant tool for a task
• Authorising execution
• Discovering the tenant's entitled catalog

How to call

POST /v1/execute (requires authentication)

{
  "skillKey": "agent.tool_policy_filter",
  "input": {
    "toolMetadata": [
      {
        "key": "mail.send_summary",
        "requiredScopes": [
          "mail:send"
        ],
        "sideEffect": "EXTERNAL_WRITE"
      },
      {
        "key": "text.build_summary",
        "requiredScopes": [
          "skills:invoke"
        ],
        "sideEffect": "NONE"
      }
    ],
    "constraints": {
      "allowedScopes": [
        "skills:invoke"
      ],
      "maxSideEffect": "NONE"
    },
    "includeExcluded": true
  }
}

Examples

{
  "toolMetadata": [
    {
      "key": "mail.send_summary",
      "requiredScopes": [
        "mail:send"
      ],
      "sideEffect": "EXTERNAL_WRITE"
    },
    {
      "key": "text.build_summary",
      "requiredScopes": [
        "skills:invoke"
      ],
      "sideEffect": "NONE"
    }
  ],
  "constraints": {
    "allowedScopes": [
      "skills:invoke"
    ],
    "maxSideEffect": "NONE"
  },
  "includeExcluded": true
}

{
  "tools": [
    {
      "key": "text.build_summary"
    }
  ],
  "summary": {
    "excludedCount": 1,
    "includedCount": 1,
    "truncated": false,
    "constraintsApplied": [
      "allowedScopes",
      "maxSideEffect"
    ],
    "candidateCount": 2
  },
  "reasons": [
    {
      "decision": "INCLUDED",
      "key": "text.build_summary",
      "reasonCode": "POLICY_ALLOWED"
    },
    {
      "decision": "EXCLUDED_BY_POLICY",
      "key": "mail.send_summary",
      "ruleId": "constraints.allowedScopes",
      "reasonCode": "SCOPE_NOT_ALLOWED"
    }
  ]
}

{
  "toolMetadata": [
    {
      "category": "text",
      "key": "z.last"
    },
    {
      "category": "text",
      "key": "a.first"
    }
  ]
}

{
  "tools": [
    {
      "category": "text",
      "key": "z.last"
    },
    {
      "category": "text",
      "key": "a.first"
    }
  ],
  "summary": {
    "excludedCount": 0,
    "includedCount": 2,
    "truncated": false,
    "constraintsApplied": [],
    "candidateCount": 2
  },
  "reasons": [
    {
      "decision": "INCLUDED",
      "key": "z.last",
      "reasonCode": "POLICY_ALLOWED"
    },
    {
      "decision": "INCLUDED",
      "key": "a.first",
      "reasonCode": "POLICY_ALLOWED"
    }
  ]
}

Recommended use cases

• Remove tools whose required scopes are outside an allowed set
• Apply side-effect, latency or cost ceilings before constructing a prompt
• Explain which explicit policy rule excluded a supplied tool

Avoid for

• Choosing the most relevant tool for a task
• Authorising execution
• Discovering the tenant's entitled catalog

Related skills

Composable with

Safety & privacy

• Runs without calling any external service.
• Your payloads are never stored permanently.
• Telemetry captures metadata only — never your payload contents.
• We never train models on your payloads.
• No human reviews your payloads.

Tags