Original templates from CLSkills Hub for coding, editing and decisions. Replace the brackets with your task. These are examples of more specific briefs, not measured model transcripts or guaranteed improvements. Use your existing Claude access.
A change is described as done, but you cannot tell what was checked.
Vague brief: Fix this bug and make sure it works.
Anyone who has used an LLM for code has seen this. You ask for a fix. You get a clean diff, a confident paragraph, and a closing line like "all tests pass." You run the tests. Two of them do not exist.
The model did not lie in any deliberate sense. The instruction was "fix this bug and make sure it works," and "make sure it works" is not an instruction. It is a wish. The model produced the sentence that a completed task would end with, because that is what completed tasks in its training data end with.
The fix I have settled on is not a smarter model or a longer system prompt. It is a short brief that changes what counts as done. I use it on real tasks with Claude (Opus 4.6 and Sonnet 4.5), in both agentic and plain chat settings. Here it is in full.
A solo founder's writeup of leaving managed hosting for a self-managed VPS, with the actual technical steps, the bugs that showed up, and the cost / performance numbers seven days in.
By Samarth Bhamare. April 2026.
A research note on which Claude "secret prompts" (L99, /skeptic, GODMODE, ULTRATHINK, and 36 others) change what Claude thinks versus what it sounds like.
Author: Samarth Bhamare. March-April 2026. Single-rater, self-funded. 15,000-word companion report as PDF link at the bottom. Raw results by code are in the appendix.
100 prompt shortcuts that change how Claude responds. Type any code at the start of your message.
Interactive searchable version with copy-to-clipboard: clskills.in/prompts
| Code | What it does |
|---|
