About improving-prompts
A skill that shortens prompts by grouping related instructions into higher-level rules while preserving their intent. It maps each original instruction to the revised prompt and flags conflicts rather than silently dropping them.
Key features
- 01Inventories prompt instructions
- 02Groups related constraints
- 03Maps intent to revised text
- 04Flags conflicts and obsolete items
- 05Prints the revision or edits on request
Use cases
- 01Condensing a long prompt
- 02Checking that a rewrite preserves requirements
- 03Reviewing conflicting prompt instructions
Install
Copy the command and run it in a terminal.
npx skills add https:/ / github.com/ bostonaholic/ skills --skill improving-promptsAfter installing, type/improving-promptsin Claude Code to confirm it loaded.
Bundled filesagents/openai.yaml
Details
From the project READMEImprove Prompt Using Concept Elevation
Concept elevation finds the higher-level rule that expresses the sum of several connected instructions. The result is shorter and lets the model adapt to new situations instead of relying on specific cases.
The argument is prompt text or a path to a file containing it. Print the improved prompt by default. Edit the file in place only when the user asks.
Process
- Number every instruction, constraint, and example in the original (I1, I2, ...). This is the intent inventory.
- Group items that serve one purpose, and replace each group with the single rule that captures all of it. Keep an item as written when no rule captures it without loss.
- Map every inventory item to the draft line that carries its intent. Restore each unmapped item, then map again, until every item maps. The draft must be shorter than the original and leave no case unaddressed that the original covered.
Never drop an item silently. When two items conflict or one looks obsolete, keep both in the draft and flag the conflict.
Deliverable
End with this format (adapt the list labels; keep the order):
```text
<the improved prompt>
```
- Merged: I1, I4, I7 -> <the rule that replaces them, abbreviated>
- Kept as written: I3 (<why no rule captured it>)
- Flagged: I9 conflicts with I2 (<the decision the user must make>)