prompt-library

Dashboard

Cost dashboards tell you what you spent. This tells you how you prompt — it reads your Claude Code session logs and surfaces the prompts that unlock autonomous work and the topics where you end up fighting the agent.

Total prompts
7,392

user turns you've sent (system messages excluded)

Sessions
138

distinct Claude Code sessions

Avg convergence
90.9%

% of turns that didn't trigger a correction

Golden prompts
536

prompts followed by ≥10 uninterrupted autonomous turns

Topics where you fight the agent most

Clusters of similar prompts with high correction rates.

Troubleshoot qBittorrent seeding and connection issues
26% correction rate
Troubleshoot tmux session recovery and display
12% correction rate
Generate and integrate visual assets
30% correction rate
Troubleshoot Instagram connector and data collection
15% correction rate
Troubleshoot Tabby and SillyTavern integration
33% correction rate
Recent prompts that just worked

Things you said right before the agent ran autonomously for a long stretch.

code/data-connectabout 23 hours ago

ultrathink through how to decomplect and then make the complete plan and then execute it if gemini approves

code/workshopabout 23 hours ago

make sure readme is strong and accurate and clear

applications/SillyTavernabout 23 hours ago

re: pass 3 the qui thing should be fixed the way other secrets work in .env, the Cap repo has some custom build stuff check docker compose to understand. i think you actually can do everything, do it

applications/SillyTavern1 day ago

<blob>

code/data-connect1 day ago

get wrong.") ⎿  2026-04-04T20:03:04.191470Z ERROR codex_core::codex: failed to load skill