WILD CORPUS · reddit
PQS 28 (F) - prompt from www.reddit.com
Source: www.reddit.com · Scraped 2026-05-04 · Scored 2026-05-04
Score
F28 / 80
gemma4:latest · local · pqs-v2.0 · canonical
Clarity7 / 10
Specificity6 / 10
Context7 / 10
Constraints3 / 10
Output format2 / 10
Role definition1 / 10
Examples1 / 10
CoT structure1 / 10
The prompt
I run a small one-person research lab where i test claude prompt codes and skill files. ran a stupid amount of tests for a guide i was writing. three things genuinely changed my workflow. 1. where you put a scope sentence matters more than how you word it i used to spend ages tweaking the wording of "review only the database logic" type sentences. then i ran the same wording at the start of a prompt vs at the end and the difference was huge. token output stayed about 30% tighter when scope went first. felt dumb for not testing it sooner. 2. about half the famous claude prompt codes do nothing tested around 120 popular codes against a fixed task battery and 47% had no measurable effect over a plain prompt. "take a deep breath" was real on older claude, doesn't reproduce on sonnet 4.6 or opus 4.7. "you are a stanford-trained expert" actually flips negative on reasoning tasks. most "step by step" variants are the default behavior already. you're typing extra characters for no reason. 3. skill file descriptions are everything if a skill .md file in \~/.claude/skills/ isn't auto-activating, the description field is almost always why. "helps with database stuff" never triggers. "use when configuring database connection pooling, choosing pool sizes, or debugging connection exhaustion" triggers reliably. vague descriptions match weakly. specific ones win. full guide is free at [Guide](http://clskillshub.com/guide) if you want the other 9 lessons. 40 pages, instant download, no email signup.
This prompt was scraped from a public source. The score reflects the input as written, not the quality of any output it produced. The AI input quality problem is the gap between what people type and what the model can act on.