When your test suite turns red before breakfast, the wrong mod leaves you with advice when you need an action. This catalogue covers 348,758 mods. Their trigger decides whether the agent reads, waits, or runs. Start with the event that matters, then choose the kind that can answer it.
Pick the trigger before the format
Ask what should cause the work. A skill is a folder of instructions the model opens when it decides it needs them. Only its name and description sit there until then. The context is everything the model can see while it answers you. A skill suits a method the agent can recognise, with a 58-token median standing cost.
A command is a named action you ask the agent to run. It suits a deliberate request. An agent suits work that deserves a separate context window. A rule fits a standing preference, especially when files match its scope. A glob is a filename pattern, like *.tsx. A matching glob lets the rule wait for relevant files instead of loading its full text each time. Commands have a 17-token median before you invoke them.
The standing price of each mod kind
A token is the unit your AI plan bills by, about three quarters of a word.
Skills add a median of 58 tokens every session. Commands add 17, agents 43, and rules 218 when always present. 90% of always-loaded mods stay at or below 142 tokens. The largest is cursorrules, at 36,510 tokens. That is closer to a long document than a quick instruction.
At 20 sessions a day, the 58-token skill median becomes 1,160 tokens. An on-demand body costs a median 710 tokens when loaded, compared with 1,206 for any mod body. Hooks themselves add no tokens, but printed output can enter the conversation.
What the scan says about risk
The scan graded 332,509 mods A, 8,968 B, 7,267 C or D, and 14 E or F. Network calls appeared in 21,506 mods, root access in 3,437, and remote-code download and execution in 2,057.
core shows a network-call finding. The C-grade parallel-cli has both network calls and remote-code execution. Each finding gives you the line that triggered it. Treat it as evidence to inspect, not a judgement of usefulness.
Five starting points for real jobs
The original is the first copy of a file we saw; everything matching it afterwards is a copy.
- skill-creator: a skill. Guidance for writing a skill of your own, for Gemini CLI. 46 tokens every session, 520 lines, original, 3 copies.
- checklist: a command. Turns a feature description into questions about what its specification leaves open. 13 tokens every session, 380 lines, original, 73 copies.
- ponytail: a rule. Tells the agent to take the simplest solution that works and reuse what is already there. 576 tokens every session, 37 lines, original, 8 copies.
- core: a skill. Controls a browser for navigation, forms, snapshots and page data. 112 tokens every session, 541 lines, original, 3 copies.
- AVM Owner Triage: a subagent. Sorts open issues across Azure Verified Modules repositories for one owner. 61 tokens every session, 683 lines, original, 2 copies.
When an event needs a hook, not a skill
A hook is a shell command your agent runs for you when something happens. Use it for session start, a tool call, a stop event, or every prompt. It alone can execute a program on your machine. A skill cannot fire on an event, though it may tell the model to run a script. A command fits a deliberate request; a rule fits a standing preference.
Questions people ask
Should I use a skill or a command? Use a skill when the agent should recognise the need, and a command when you will name the action yourself. The command median is 17 tokens, versus 58 for a skill.
Is a skill or an agent better for a long job? Choose an agent when a self-contained result can return to your main work. Choose a skill when the main agent must apply the instructions as it works. The agent median is 43 tokens.
When should I use a hook? Choose one when an event must start the action without a request. A session-start hook that prints 1,296 tokens adds that output to the conversation every session, so keep automatic output short.
Do it now
Run npx agentmods cost to see per-session costs in your project, then browse all mods.