AI Skill Report Card
Upgrading Prompts
Upgrades a prompt file to a target sophistication level (1-7), adding the structural sections and patterns appropriate to that level.
Quick Start12 / 15
Upgrade prompt.md to level 4
Reads prompt.md, detects its current level (e.g., level 2), and adds the missing sections needed to reach level 4 (Control Flow + Delegation), then writes the result back and reports the diff.
If no target level is given, default to current level + 1.
Recommendation▾
Add a second example showing a smaller single-level upgrade (e.g., 1→2) and one showing a multi-level jump warning in action, to demonstrate range.
The Seven Levels
| Level | Name | Key Sections/Patterns |
|---|---|---|
| 1 | Basic | Plain instructions |
| 2 | Structured | Variables, numbered Workflow, Report |
| 3 | Control Flow | STOP conditions, <loop> tags, conditionals |
| 4 | Delegation | Task tool calls, parallel agents, result aggregation |
| 5 | Meta | Accepts a prompt file path as input, higher-order processing |
| 6 | Generative | Specified Format templates, meta-prompt generation, doc fetching |
| 7 | Expert | Expertise section, self-improvement loop, knowledge accumulation |
Workflow14 / 15
Progress:
- Step 1: Validate inputs (prompt file path required; stop and ask if missing)
- Step 2: Read the full prompt file
- Step 3: Detect current level from sections/patterns present
- Step 4: Determine target level (arg or current+1); validate target > current and ≤ 7; warn if jump > 2
- Step 5: Map required additions for each level transition in the jump
- Step 6: Transform the prompt, adding sections in order without removing existing content
- Step 7: Save file and output upgrade report
Step 3: Level Detection Heuristics
- Has "Variables" + numbered "Workflow" + "Report" → at least level 2
- Has STOP conditions or
<loop>tags or if/else branching → at least level 3 - Has Task tool invocations or "launch agents in parallel" → at least level 4
- Accepts a prompt/file path as its primary input and processes it → at least level 5
- Has a "Specified Format" / template-generation section → at least level 6
- Has "Expertise" section or self-improvement/knowledge-accumulation loop → level 7
Step 5: Addition Map (apply cumulatively for multi-level jumps)
| Transition | Add |
|---|---|
| 1→2 | Variables section (dynamic/static), numbered Workflow, Report section |
| 2→3 | STOP conditions in Workflow, <loop> tags for iteration, conditional branches |
| 3→4 | Task tool delegation, parallel agent launch, result aggregation step |
| 4→5 | PATH_TO_PROMPT variable, logic to read/process another prompt, higher-order framing |
| 5→6 | Specified Format template block, meta-prompt generation workflow, doc fetching step |
| 6→7 | Expertise section, self-improvement workflow, knowledge accumulation pattern |
Step 6: Transformation Rules
- Never delete existing content — only add or restructure.
- Insert new sections in canonical order: Variables → Workflow → Control Flow → Delegation → Report → Specified Format → Expertise.
- Preserve original variable names and instructions; wrap them, don't rewrite intent.
- Make STOP conditions explicit and imperative ("If X, STOP and ask user for Y").
Recommendation▾
Show a concrete before/after diff of actual prompt text (not just a report template) so the transformation mechanics are verifiable rather than described abstractly.
Example
Input: prompt.md at level 2 (has Variables + Workflow + Report), target level 4.
Output:
Markdownundefined
Prompt Upgraded
File: prompt.md From Level: 2 (Structured) To Level: 4 (Delegation)
Changes Made
Sections Added:
- Control Flow (STOP conditions, loop tags)
- Delegation (Task tool pattern, parallel agents, aggregation)
Patterns Added:
- Explicit STOP-and-ask validation step
- Parallel Task-tool agent launch for independent subtasks
- Result aggregation step before Report
New Structure
Markdown[preview of updated prompt.md]
Validation
- New sections properly formatted
- Variables follow conventions
- Workflow updated for new level
- STOP conditions explicit
Testing Recommendations
- Test basic functionality unchanged from level 2
- Test new control-flow branches and STOP triggers
- Test delegation with real parallel Task calls
- Verify final Report output format
Best Practices
- Apply the 80/20 rule: levels 3-4 (Control Flow, Delegation) cover most real use cases — don't push to 6-7 without a clear need.
- Upgrade only what's needed; a jump of more than 2 levels should trigger a warning about added complexity.
- Keep the original prompt's voice and domain content intact — you're adding scaffolding, not rewriting purpose.
- Always show a before/after preview so the user can verify the transformation.
Common Pitfalls
- Don't skip level detection and assume level 1 — misdetecting current level causes duplicate or conflicting sections.
- Don't remove or "clean up" existing instructions while upgrading — this breaks trust in the transformation.
- Don't add Delegation (Task tool) patterns to prompts that have no independent subtasks to parallelize — it adds complexity without benefit.
- Don't jump straight to level 7 (Expertise/self-improvement) unless the prompt is genuinely long-running and knowledge-accumulating — most prompts don't need it.