your-skill-name

name: your-skill-name description: >- Brief description of what this skill does and what tools/integrations it uses. Use when: (1) first trigger scenario, (2) second trigger scenario, (3) third trigger scenario.

Your Skill Name

One-line summary of the skill's purpose and outcome.

Success target: [Measurable success criteria — e.g., "90%+ of X produce correct Y."]

Task

[High-level pipeline in one line: Step A → Step B → Step C → Return result.]

1. Accept Input

Input Formats

Format A (primary):

• field_1: Description
• field_2: Description
• field_3: Description (optional)

Format B (fallback): Any raw text block to analyze.

If only raw text is provided, treat it as the full content to analyze.

2. Classify / Analyze

Categories

Classification Process

Step 1: Scan for indicators

Read the input and identify:

• [Signal type 1]
• [Signal type 2]
• [Signal type 3]

Step 2: Match categories

Compare indicators against the categories above. A single input may match multiple categories.

Step 3: Extract structured data per match

For each match, extract:

• category: The matched category
• key_field_1: Extracted value
• key_field_2: Extracted value
• deadline: Date if found (YYYY-MM-DD)

Step 4: Calculate confidence score

Each match gets an independent score. Use the point-based system below.

Confidence Scoring System

Start with confidence = 0.0 and add points:

Maximum possible: 1.0

Important: All example breakdowns in this skill must sum correctly. Verify: 0.30 + 0.25 + 0.20 + 0.15 + 0.10 = 1.0 ✓

Examples:

• "[Full match example]" → 0.30 + 0.25 + 0.20 + 0.15 + 0.10 = 1.0
• "[Partial match example]" → 0.20 = 0.20 (below threshold)

What to Ask If Missing

Classification Output

{
  "matches": [
    {
      "category": "category_a",
      "confidence": 1.0,
      "confidence_breakdown": {
        "signal_a": 0.30,
        "signal_b": 0.25,
        "signal_c": 0.20,
        "signal_d": 0.15,
        "signal_e": 0.10
      },
      "key_field_1": "extracted_value",
      "key_field_2": "extracted_value",
      "priority": "Medium",
      "suggested_title": "Action: Category — Context — Entity",
      "notes": "Brief description.",
      "missing_signals": []
    }
  ]
}

Confidence Thresholds

3. Decision Logic

1. Classify input → get matches[]
2. If matches[] is empty → inform user, done
3. For each match in matches[]:
   a. confidence >= 0.85 → Execute action (auto)
   b. confidence >= 0.45 and < 0.85 → Execute action with review flag
   c. confidence < 0.45 → Skip, note below threshold
4. On failure for one match → Log error, continue with remaining
5. After all processed → Return aggregated response (see Section 5)

Each match is independent. A failure on one does not block the others.

4. Execute Action

[Describe the action — e.g., create a record, call an API, generate output.]

Required Tools / Integrations

For tool schemas, field definitions, and detailed reference material, read REFERENCE.md.

5. Response to User

Success (one or more actions taken)

[N] action(s) completed from this input.

For each high-confidence action:

[n]. Title: [title]
     Status: [status]
     Priority: [priority]
     Record: [id]

For each mid-confidence action:

[n]. Title: [title]
     Status: Pending Review
     Priority: [priority]
     Record: [id]
     Flagged for review — confidence was [X]%.

If any matches were skipped, append:

Skipped: [category] — confidence [X]% below 45% threshold.

If any actions failed, append:

Failed: [category] — [error message]. Please handle manually.

No Matches Found

Analyzed input — no [domain] action detected.

No actions were taken. [Brief explanation.]

All Below Threshold

Possible [domain] matches detected, but confidence too low.

[category]: [X]% (below 45% threshold)

No actions taken. Provide more context if action is needed.

6. Mapping Tables

Tag / Category Mapping

Priority Mapping

Handling Errors

Example

For a complete worked example with input, classification, and output, read EXAMPLE.md.