# Prompt Decomposition Engine (PDE) — llms-txt Reference

> Intention exploration as EAST practice: decomposing prompts into explicit and implicit intents, mapping dependencies through Four Directions, and holding ambiguities as productive tension.

**Version**: 0.1 (stub — evolving from practice)
**Document ID**: llms-pde-v0.1
**Last Updated**: 2026-02-27
**Content Source**: Derived from practical PDE usage in this repository and the medicine-wheel-prompt-decomposition npm package.

---

## Purpose

PDE transforms a complex prompt into a structured decomposition that:

1. Identifies **primary** and **secondary** intents (explicit and implicit)
2. Maps each intent to a **Four Directions** position (East/South/West/North)
3. Surfaces **ambiguities** as named tensions — not silently resolved assumptions
4. Produces an **action stack** respecting dependencies between tasks

---

## Four Directions in PDE

The standard PDE directions (adapted from creative-orientation practice):

| Direction | Role in Decomposition | Phase |
|-----------|----------------------|-------|
| 🌅 **East** — Vision | Understanding what is being asked; clarifying requirements; envisioning desired outcomes | Germination |
| 🔥 **South** — Analysis | Research, learning, investigation, growth tasks; planning with human and AI companions | Assimilation |
| 🌊 **West** — Validation | Testing, reflection, review, accountability checks | Assimilation |
| ❄️ **North** — Action | Implementation, execution, delivery, completion | Completion |

**Note**: These directions are a starting point. The creative-orientation llms-txt suggests they may be adjusted as practice evolves. PDE is itself an EAST practice — the decomposition is part of the visioning phase where intentions are explored before action begins.

---

## Decomposition Structure

```json
{
  "primary": {
    "action": "main action verb",
    "target": "what the action applies to",
    "urgency": "immediate|session|persistent",
    "confidence": 0.0-1.0
  },
  "secondary": [
    {
      "action": "verb",
      "target": "target",
      "implicit": true/false,
      "dependency": "what this depends on",
      "confidence": 0.0-1.0
    }
  ],
  "directions": {
    "east": [{"text": "vision items", "confidence": 0.0-1.0, "implicit": false}],
    "south": [{"text": "analysis items"}],
    "west": [{"text": "validation items"}],
    "north": [{"text": "action items"}]
  },
  "actionStack": [
    {"text": "task", "direction": "east|south|west|north", "dependency": "or null"}
  ],
  "ambiguities": [
    {"text": "ambiguous part", "suggestion": "how to clarify"}
  ]
}
```

---

## Key Principles

### Implicit Intent Extraction

PDE extracts implicit intents from hedging language: "I assume," "probably," "somehow," "you will need," "I expect." These are flagged with `"implicit": true` and lower confidence scores.

### Delayed Resolution of Ambiguities

Ambiguities are **named and surfaced**, not silently resolved. This honors the delayed-resolution principle — holding productive tension rather than collapsing it through premature assumptions.

### Confidence Scoring

Each intent carries a confidence score (0.0–1.0) reflecting how clearly it was stated. Explicit requests score high; implicit inferences score lower. This transparency enables the human steward to adjust course.

---

## PDE as Relational Practice

When used with an AI companion, PDE is a **ceremonial act of shared visioning**:

- The human brings the intention (however messy or complex)
- The AI companion surfaces what was said and what was implied
- Together they refine the vision before entering SOUTH (planning) and NORTH (action)
- Ambiguities become conversation starters, not blockers

---

## Storage

Decompositions are stored in `.pde/` as paired JSON and Markdown files:
- `{uuid}.json` — machine-readable structured decomposition
- `{uuid}.md` — human-editable Four Directions document

The markdown file is designed for human annotation — the steward can edit, add notes, correct interpretations, and refine the vision before work begins.

---

## Related

- [medicine-wheel-prompt-decomposition](https://www.npmjs.com/package/medicine-wheel-prompt-decomposition) — npm package
- [llms-creative-orientation.txt](llms-creative-orientation.txt) — foundational framework
- [llms-delayed-resolution-principle.md](llms-delayed-resolution-principle.md) — tension maintenance
- [llms-structural-tension-charts.txt](llms-structural-tension-charts.txt) — charting methodology
- [docs/prompt-decomposition.md](docs/prompt-decomposition.md) — extended documentation