Approving one tool request also determines what that permission can open. Settings set the reach, so choose the permission you can explain, not the file with the friendliest name. The catalogue measures 2,888 settings, making this boundary worth checking before installation.
What a settings file changes
Settings are configuration, not text the model reads. They control which tools may run without asking first. A glob is a filename pattern, like *.tsx. A permission using one can cover every matching file, including files you did not have in mind.
A hook is a shell command your agent runs for you when something happens. It can run at session start or around tool use. A permission written for one command allows every use of that command, including uses you did not have in mind.
Claude Code is the largest catalogue subject, with 752 settings. A settings file of 16 lines can change what your agent is allowed to do. Read the allowed action before trusting the filename.
No token cost, and what the file changes instead
Settings cost nothing per session. A token is the unit your AI plan bills by, about three quarters of a word. There is therefore no median or 90th-percentile token cost, and no maximum-cost settings mod to name.
File length is not token spend here: the files below run from 16 lines to 269, and all of them cost the same nothing.
What the settings scan puts in view
The scan grades 2,580 A, 53 B, 252 C or D, and 0 E or F. An A means it found nothing it treats as risky. It does not mean the page lists no finding.
Recursive force delete appears in 234 settings. Network calls appear in 189, requests for root in 72, and credential-file access in 17. Eleven run shell commands, 3 enumerate installed skills, and 2 download and execute remote code.
The ccxt setting is graded C for recursive force delete. The claude-code-game-studios setting is graded D for asking for root and recursive force delete. The mockserver-monorepo setting is graded C for recursive force delete.
The scan prints the line that produced each finding. Read that line when you review the mod. The scan does not judge the mod for you.
Five settings worth starting with
The original is the first copy of a file we saw; everything matching it afterwards is a copy.
- supabase: blocks protected file changes and runs setup, formatting, and linting commands. 38 lines, original, 0 copies, 0 tokens every session.
- superset: runs pre-commit checks before the agent creates a Git commit. 16 lines, original, 0 copies, 0 tokens every session.
- RuView: runs commands around prompts, tools, and sessions for routing, state restoration, and edit checks. 269 lines, original, 0 copies, 0 tokens every session.
- orm: controls allowed tools and scripts before commands, after edits, worktree creation, and stop. 65 lines, original, 1 copy, 0 tokens every session.
- claude-cookbooks: starts environment checks and warns before protected writes or risky shell commands. 31 lines, original, 1 copy, 0 tokens every session.
When a rules or skill mod fits better
Choose a rules or skill mod when the job is to tell the model what to do, explain project context, or shape its style. A skill is a folder of instructions the model opens when it decides it needs them. Settings cannot provide that guidance. They configure permissions and automatic actions instead. The catalogue tags 49 of these files for memory and context.
Questions people ask
What does a *.tsx permission really allow? It covers every matching filename, not only the file you had in mind. Decide whether you want read, write, or execute access before accepting the pattern.
Is an A grade permission automatically safe? Use it as a first filter, not permission to skip reading. Only 39 mods have been reviewed, and none has high review risk, so your own scope check still matters.
How many allowed tools should make me pause? There is no universal limit. One catalogue setting reaches 25 allowed tools; use 25 as a review prompt, not a safety guarantee.
Do it now
Run npx agentmods add <slug> to install one mod by its catalogue slug, then browse settings.