@agent-pattern-labs/iso-facts

Deterministic fact materialization for AI-agent workflows.

iso-facts extracts structured, provenance-rich facts from local artifacts without model calls. It is the layer between "where is the source file?" (iso-index) and "is the downstream candidate record valid?" (iso-contract / iso-preflight).

Install

npm install @agent-pattern-labs/iso-facts

CLI

iso-facts build --config facts.json --root . --out .iso-facts.json
iso-facts query --facts .iso-facts.json --fact job.url
iso-facts has --facts .iso-facts.json --fact job.score
iso-facts verify --facts .iso-facts.json
iso-facts check --facts .iso-facts.json --config facts.json
iso-facts explain --config facts.json

Config

{
  "version": 1,
  "sources": [
    {
      "name": "reports",
      "include": ["reports/*.md"],
      "format": "text",
      "rules": [
        {
          "fact": "job.url",
          "pattern": "^\\*\\*URL:\\*\\*\\s*(?<url>https?://\\S+)",
          "flags": "i",
          "key": "{url}",
          "value": "{url}",
          "fields": {
            "url": "{url}",
            "report": "{source}"
          },
          "tags": ["report", "url"]
        }
      ]
    }
  ],
  "requirements": [
    { "fact": "job.url", "min": 1 }
  ]
}

Supported source formats:

• text: regex rules run line-by-line.
• tsv: rows become structured inputs; headers are read from the first row by default.
• markdown-table: markdown rows become structured inputs.
• jsonl: each JSON object line becomes a structured input.
• json: each configured path selects one or more structured inputs.

Template fields support {field} placeholders and filters:

• trim
• lower
• upper
• slug
• json

Example: {company|slug}:{role|slug}.

Fact Set

build writes a deterministic fact set:

{
  "schemaVersion": 1,
  "root": "/repo",
  "configHash": "...",
  "stats": { "sources": 1, "files": 1, "facts": 1 },
  "facts": [
    {
      "schemaVersion": 1,
      "id": "...",
      "fact": "job.url",
      "key": "https://jobs.example.com/staff-ai-engineer",
      "value": "https://jobs.example.com/staff-ai-engineer",
      "source": {
        "name": "reports",
        "path": "reports/001-example.md",
        "line": 3
      },
      "fields": {
        "url": "https://jobs.example.com/staff-ai-engineer"
      },
      "tags": ["report", "url"]
    }
  ]
}

Findings preserve source path and line, so downstream agents can load only the authoritative source span instead of rereading broad artifact trees.

Library

import {
  buildFacts,
  checkFactRequirements,
  hasFact,
  loadFactsConfig,
  queryFacts,
  verifyFactSet,
} from "@agent-pattern-labs/iso-facts";

const config = loadFactsConfig(JSON.parse(await fs.readFile("facts.json", "utf8")));
const factSet = buildFacts(config, { root: process.cwd() });

if (hasFact(factSet, { fact: "job.url" })) {
  console.log(queryFacts(factSet, { fact: "job.url" }));
}

console.log(verifyFactSet(factSet));
console.log(checkFactRequirements(factSet, config.requirements));

Boundaries

iso-facts does not decide source precedence, identity matching, artifact shape, dispatch safety, or retention policy.

• Use iso-index to find compact source pointers.
• Use iso-canon for identity keys and comparisons.
• Use iso-contract for record validation/rendering.
• Use iso-preflight to turn materialized facts into safe dispatch plans.
• Use iso-redact before exporting fact inputs or fact sets that may contain sensitive data.