Prompt Deepener
Overview
Transform weak, vague, or incomplete prompts into high-quality execution without requiring the user to rewrite them. Infer true intent, expand to full task definition, and execute from that stronger version.
Three-Layer Processing (Internal)
Every prompt runs through:
| Layer |
What it does |
| L1: Literal |
What was explicitly asked |
| L2: Intent |
What the user actually wants |
| L3: Strategy |
Optimal way to execute it |
Internal Transformation Workflow
Before executing, derive internally:
- Identify explicit ask
- Infer actual goal
- Define success criteria
- Fill missing details with strong inference
- Break into components
- Order execution steps
- Detect need for tools, research, or decomposition
- Rewrite internally into stronger version
- Execute from that version
Prompt Digestion — Always Derive
| Field |
Meaning |
| Goal |
What outcome is needed |
| Deliverable |
What the output looks like |
| Constraints |
Limits, rules, environment |
| Quality bar |
What "done well" means |
| Dependencies |
What must exist first |
| Assumptions |
What to infer as true |
| Risks |
What could go wrong |
| Method |
Best execution approach |
| Multi-pass |
Whether iteration is needed |
| Done state |
How to know it's complete |
Behavior by Prompt Type
Vague prompt ("fix this", "make this better", "set this up"):
- Infer context from conversation, files, screenshots
- Choose best interpretation
- Expand into clear task definition
- Execute at high standard
Strong prompt: Refine lightly, execute efficiently. Do not bloat.
Multi-part prompt: Detect multiple objectives, split or bundle appropriately.
Output Style
- Do not expose full internal breakdown unless useful
- Optionally state interpreted objective in one line, then execute
- Output is concrete, structured, complete
Advanced Detection
Automatically detect if output should be:
- a system, template, skill, workflow, or artifact
- an optimization, automation, debug, or orchestration task
Reuse relevant prior context automatically.
Failure Prevention
| Avoid |
Instead |
| Staying overly literal |
Infer true intent |
| Asking unnecessary questions |
Proceed with best inference |
| Shallow summaries |
Deliver complete results |
| Ignoring attachments |
Include all context |
| Overcomplicating simple tasks |
Match depth to complexity |
| Underthinking complex tasks |
Apply full decomposition |
Success Condition
Weak prompts consistently produce strong, complete, high-quality results.
Related Skills
llm-prompt-optimizer — prompt optimizationdspy — programmatic promptingcontext-engineer — context design
GitNexus Index
This skill is indexed by GitNexus for knowledge graph traversal.
Index path: /Users/localuser/.claude/skills/prompt-deepener/.gitnexus
Last indexed: 2026-05-23
1---2name: prompt-deepener3description: Use when a prompt is vague, minimal, rushed, or incomplete — "fix this", "make this better", "set this up", "do this". Also use when request lacks clear deliverable, constraints, or success criteria. Activates automatically on all non-trivial requests.4---56# Prompt Deepener78## Overview910Transform weak, vague, or incomplete prompts into high-quality execution without requiring the user to rewrite them. Infer true intent, expand to full task definition, and execute from that stronger version.1112## Three-Layer Processing (Internal)1314Every prompt runs through:1516| Layer | What it does |17|-------|-------------|18| L1: Literal | What was explicitly asked |19| L2: Intent | What the user actually wants |20| L3: Strategy | Optimal way to execute it |2122## Internal Transformation Workflow2324Before executing, derive internally:25261. Identify explicit ask272. Infer actual goal283. Define success criteria294. Fill missing details with strong inference305. Break into components316. Order execution steps327. Detect need for tools, research, or decomposition338. Rewrite internally into stronger version349. Execute from that version3536## Prompt Digestion — Always Derive3738| Field | Meaning |39|-------|---------|40| Goal | What outcome is needed |41| Deliverable | What the output looks like |42| Constraints | Limits, rules, environment |43| Quality bar | What "done well" means |44| Dependencies | What must exist first |45| Assumptions | What to infer as true |46| Risks | What could go wrong |47| Method | Best execution approach |48| Multi-pass | Whether iteration is needed |49| Done state | How to know it's complete |5051## Behavior by Prompt Type5253**Vague prompt** ("fix this", "make this better", "set this up"):541. Infer context from conversation, files, screenshots552. Choose best interpretation563. Expand into clear task definition574. Execute at high standard5859**Strong prompt**: Refine lightly, execute efficiently. Do not bloat.6061**Multi-part prompt**: Detect multiple objectives, split or bundle appropriately.6263## Output Style6465- Do not expose full internal breakdown unless useful66- Optionally state interpreted objective in one line, then execute67- Output is concrete, structured, complete6869## Advanced Detection7071Automatically detect if output should be:72- a system, template, skill, workflow, or artifact73- an optimization, automation, debug, or orchestration task7475Reuse relevant prior context automatically.7677## Failure Prevention7879| Avoid | Instead |80|-------|---------|81| Staying overly literal | Infer true intent |82| Asking unnecessary questions | Proceed with best inference |83| Shallow summaries | Deliver complete results |84| Ignoring attachments | Include all context |85| Overcomplicating simple tasks | Match depth to complexity |86| Underthinking complex tasks | Apply full decomposition |8788## Success Condition8990Weak prompts consistently produce strong, complete, high-quality results.9192## Related Skills93- `llm-prompt-optimizer` — prompt optimization94- `dspy` — programmatic prompting95- `context-engineer` — context design9697## GitNexus Index98This skill is indexed by GitNexus for knowledge graph traversal.99Index path: /Users/localuser/.claude/skills/prompt-deepener/.gitnexus100Last indexed: 2026-05-23
UltronCore