viasrijan Prompt Enhancement
Transform weak or vague prompts into precision-engineered instructions. Composite of brixtonpham/claude-config prompt-enhancer + llm-prompt-optimizer RSCIT.
When to use
- Input is vague, inconsistent, or hallucination-prone
- Need structured/JSON output reliably
- Want to cut input tokens without losing effectiveness
- This skill is always-on in Crazy Skills — also wired as
UserPromptSubmithook for auto-enhancement
Workflow
1. Diagnose
| Problem | Symptom | Fix |
|---|---|---|
| Too vague | Generic answers | Add Role + Context + Constraints |
| No structure | Unformatted | Specify output format |
| Hallucination | Confident wrong | Add "say I don't know if unsure" |
| Inconsistent | Different each run | Add few-shot examples |
| Too long | Verbose | Add length constraints |
2. Apply RSCIT
- R — Role: Who is the AI? (e.g., "senior ML engineer")
- S — Situation: What context? (e.g., "1 year Python, no ML background")
- C — Constraints: Rules and limits (e.g., "≤200 words, no formulas, use analogy")
- I — Instructions: What exactly to do?
- T — Template: What output looks like?
Before: Explain machine learning.
After: You are a senior ML engineer. Context: junior with 1 year Python. Task: explain supervised ML simply. Constraints: analogy, ≤200 words, no formulas, one next step. Format: plain prose.
3. Compress filler
Please carefully analyze the following code and provide... → Analyze this code: explain what it does, how it works, flag issues.
4. Hallucination guard
Answer based ONLY on provided context. If not contained, respond exactly: "I don't have enough information."
Verification
- Role, Situation, Constraints, Instructions, Template are explicit
- Output format is explicitly defined (JSON/markdown/bullets)
- Hallucination guard present for factual tasks
- Prompt is compressed (filler removed) without losing task-critical instructions
References
- Source:
brixtonpham/claude-configprompt-enhancer,mohamednaeem92-max/OPENCODE-6-2026llm-prompt-optimizer - Hook:
UserPromptSubmit→ prompt-enhancement (optional auto-mode inopencode.json)