Field Extractor

Extracts named fields from unstructured text, incident logs, or support tickets. Assigns explicit status (FOUND, MISSING, AMBIGUOUS) to every requested field with zero hallucination.

1. Purpose

Harvest targeted attributes from text. A "Field" is a specific named variable (e.g. ip_address, error_code, user_id).

2. When to Use

Use this skill when:

• Pulling specific parameters from server logs or crash traces.
• Extracting contact details or invoice amounts from emails.
• User requests: "Extract [fields] from this text."

3. When Not to Use

• Do not use to convert entire complex structures to validated JSON; use json-schema-converter.
• Do not use to summarize full prose text; use factual-summarizer.

4. Required Inputs

• source_document: Text or log snippet to parse.
• target_field_names: List of specific field names to extract.

5. Missing-Input Behavior

1. If source_document is missing: stop and ask: "Please provide the document to extract fields from."
2. If target_field_names is missing: ask: "Which specific fields should be extracted?".
3. If a field is not found in text: label as MISSING (Not present in text). Do not guess.

6. Procedure

Follow these numbered actions:

1. Action 1: Enumerate Target Fields
   • Create an extraction entry for every requested field.
2. Action 2: Scan Document for Exact or Contextual Matches
   • Locate regex patterns, key-value delimiters (:, =), or natural language mentions.
3. Action 3: Categorize Field Status
   • FOUND: Exact value found; include quotation proof.
   • AMBIGUOUS: Multiple conflicting candidates found; list all candidates.
   • MISSING: No candidate found in text.
4. Action 4: Emit Extraction Table
   • Format using Section 8.

7. Decision Rules

• If multiple IP addresses or dates are found and field is singular: label status as AMBIGUOUS and list each candidate with line context.
• Never guess passwords, phone numbers, or emails from missing text.

8. Output Format

Emit the extracted fields using this structured template:

### Extracted Fields Summary

| Field Name | Status | Extracted Value | Evidence / Context |
| :--- | :--- | :--- | :--- |
| `[field_1]` | **FOUND** | `[value]` | "[Context snippet]" |
| `[field_2]` | **MISSING** | *null* | *Not mentioned in source text.* |
| `[field_3]` | **AMBIGUOUS** | `[cand_1], [cand_2]` | Found 2 matches on lines 4 and 12 |

#### Extraction Statistics
- **Requested**: [Count] | **Found**: [Count] | **Missing**: [Count] | **Ambiguous**: [Count]

9. Verification Checklist

• Every requested field is present in the table.
• Missing fields are explicitly marked as null / MISSING.
• Evidence quotes confirm each FOUND value.
• No values were hallucinated.

10. Common Failures and Recovery

11. Stop or Escalation Conditions

Stop and ask the user when:

• Source document is unparseable binary.

12. Examples

Example 1: Successful Execution

Input:

Document: "Failed login attempt from IP 192.168.1.50 for user admin at 2026-09-10T08:14:22Z with error code ERR_INVALID_PASS."
Fields: ip_address, username, error_code, retry_count

Output:

### Extracted Fields Summary

| Field Name | Status | Extracted Value | Evidence / Context |
| :--- | :--- | :--- | :--- |
| `ip_address` | **FOUND** | `192.168.1.50` | "...from IP 192.168.1.50..." |
| `username` | **FOUND** | `admin` | "...for user admin..." |
| `error_code` | **FOUND** | `ERR_INVALID_PASS` | "...with error code ERR_INVALID_PASS." |
| `retry_count` | **MISSING** | *null* | *Not mentioned in source text.* |

#### Extraction Statistics
- **Requested**: 4 | **Found**: 3 | **Missing**: 1 | **Ambiguous**: 0

Example 2: Ambiguous or Missing Input

Input:

Document: "Invoice total is $500, with a discount of $50 leaving $450 due."
Fields: amount

Output:

### Extracted Fields Summary

| Field Name | Status | Extracted Value | Evidence / Context |
| :--- | :--- | :--- | :--- |
| `amount` | **AMBIGUOUS** | `$500, $50, $450` | Multiple dollar amounts found (total, discount, amount due). |

#### Extraction Statistics
- **Requested**: 1 | **Found**: 0 | **Missing**: 0 | **Ambiguous**: 1

Example 3: Expected Final Output

Output:

### Extracted Fields Summary

| Field Name | Status | Extracted Value | Evidence / Context |
| :--- | :--- | :--- | :--- |
| `order_id` | **FOUND** | `ORD-9901` | "Order reference ORD-9901 confirmed." |
| `tracking_number` | **MISSING** | *null* | *Shipment not yet initiated; no tracking number.* |

#### Extraction Statistics
- **Requested**: 2 | **Found**: 1 | **Missing**: 1 | **Ambiguous**: 0