Agent Step Result Normalizer

Overview

This advises and does not act: it reshapes a result you already have and calls nothing. Whatever a step returned — an object, an array, a bare string, a number, null — comes back as the same envelope, so an agent branches on one shape instead of one shape per tool. Status is derived by published rules for the profile you pick, and it is never guessed upwards: a payload the profile does not recognise becomes unknown, because inferring success is exactly how an agent carries on after a step that failed. Every ceiling is applied while walking the payload rather than after copying it, so a 2 MB result never becomes a 2 MB envelope, and each reduction is flagged so you know the envelope is partial. redactKeys removes values by exact field name as a convenience for names you already know; it is not a scanner, and security.detect_secrets and security.redact_secrets are the skills for that.

An agent that branches on each tool's own result shape grows a special case per tool, and the worst of those special cases is the one that reads an unfamiliar payload as success and carries on. One envelope with a status that falls back to unknown removes both problems, and bounding the payload keeps a large result from filling the context window.

When to use it

Scope

Non-goals

• Does not call the tool that produced the payload, or any other tool
• Does not scan for secrets or personal data — redactKeys matches names only
• Does not repair a malformed payload, only describe and bound it
• Does not infer success from a payload its profile cannot read
• Does not keep or cache the payload after the answer is returned

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.step_result_normalize","input":{"toolResult":{"slug":"weekly-report"},"profile":"SQUILLIFY"}}'

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

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

const { output } = await sq.execute("agent.step_result_normalize", {"toolResult":{"error":{"message":"The file reference is not accessible with the current credentials.","code":"FILE_REF_NOT_ACCESSIBLE"},"requestId":"req-8813"},"redactKeys":["requestId"],"stepId":"s-3","tool":"file.read"});

Errors

Other limits

Limitations

• Does not call the tool that produced the payload, or any other tool
• Does not scan for secrets or personal data — redactKeys matches names only
• Does not repair a malformed payload, only describe and bound it
• Does not infer success from a payload its profile cannot read
• Does not keep or cache the payload after the answer is returned
• Removing secrets or personal data before storage
• Keeping a payload intact — this deliberately bounds and drops
• Deciding what to do next, which is the agent's job

How to call

POST /v1/execute (requires authentication)

{
  "skillKey": "agent.step_result_normalize",
  "input": {
    "toolResult": {
      "error": {
        "message": "The file reference is not accessible with the current credentials.",
        "code": "FILE_REF_NOT_ACCESSIBLE"
      },
      "requestId": "req-8813"
    },
    "redactKeys": [
      "requestId"
    ],
    "stepId": "s-3",
    "tool": "file.read"
  }
}

Examples

{
  "toolResult": {
    "error": {
      "message": "The file reference is not accessible with the current credentials.",
      "code": "FILE_REF_NOT_ACCESSIBLE"
    },
    "requestId": "req-8813"
  },
  "redactKeys": [
    "requestId"
  ],
  "stepId": "s-3",
  "tool": "file.read"
}

{
  "status": "error",
  "stepId": "s-3",
  "data": {
    "error": {
      "message": "The file reference is not accessible with the current credentials.",
      "code": "FILE_REF_NOT_ACCESSIBLE"
    },
    "requestId": "<removed>"
  },
  "metrics": {
    "payloadChars": 146,
    "depth": 2,
    "itemCount": 0
  },
  "truncation": {
    "textTruncated": false,
    "itemsTruncated": false,
    "depthTruncated": false,
    "fieldsDropped": 0
  },
  "tool": "file.read",
  "items": [],
  "summary": "2 fields",
  "error": {
    "code": "FILE_REF_NOT_ACCESSIBLE",
    "message": "The file reference is not accessible with the current credentials."
  },
  "profile": "squillify"
}

{
  "toolResult": {
    "body": {
      "invoice": "inv-9912",
      "code": "NOT_FOUND"
    },
    "statusCode": 404,
    "message": "No such invoice"
  },
  "stepId": "s-7",
  "tool": "billing.fetch_invoice",
  "profile": "HTTP"
}

{
  "data": {
    "body": {
      "invoice": "inv-9912",
      "code": "NOT_FOUND"
    },
    "statusCode": 404,
    "message": "No such invoice"
  },
  "summary": "3 fields, 15 characters of text",
  "truncation": {
    "textTruncated": false,
    "itemsTruncated": false,
    "depthTruncated": false,
    "fieldsDropped": 0
  },
  "status": "client_error",
  "tool": "billing.fetch_invoice",
  "error": {
    "code": "HTTP_CLIENT_ERROR",
    "message": "No such invoice"
  },
  "stepId": "s-7",
  "text": "No such invoice",
  "profile": "http",
  "metrics": {
    "payloadChars": 95,
    "depth": 2,
    "itemCount": 0
  },
  "items": []
}

Recommended use cases

• Give an agent loop one result shape to branch on across many tools
• Cut a multi-megabyte tool result down to something a prompt can hold
• Turn a raw HTTP response into a status an agent can act on

Avoid for

• Removing secrets or personal data before storage
• Keeping a payload intact — this deliberately bounds and drops
• Deciding what to do next, which is the agent's job

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