When your agent turns a rough request into code, the plan decides what happens before the first edit. Choose the plan that resolves the uncertainty blocking you, rather than adding instructions you will not use. With this many options, cost and fit both matter.
What sits in the session before you plan anything
A skill is a folder of instructions the model opens when it decides it needs them. The context is everything the model can see while it answers you.
The catalogue contains 26,049 planning skills. Claude Code accounts for 24,434 entries, Codex for 11,987, and Cursor for 177. A skill can be labelled for more than one of them, so choose for the agent you use and the decision you need settled.
Use brainstorming when the idea is unclear. Use spec_generator when the issue is known but the affected files and work steps need structure. Use a larger method when the work has several stages or will not fit in one session.
The 56-token preload
A token is the unit your AI plan bills by, about three quarters of a word. The median planning skill costs 56 tokens every session, about a short email’s worth of setup. Only the top tenth passes 138 tokens.
At 20 sessions a day, the 56-token median becomes 1,120 tokens across a day. The largest listed preload is finesse-brief, at 1,267 tokens every session across 660 lines. The median full body for an on-demand skill is 1,136 tokens when loaded.
What the safety scan puts on the page
The scan graded 25,341 skills A, 421 B, and 287 C or D. It found 434 skills making network calls, 202 reading agent configuration directories, and 118 using recursive force delete.
The understand skill is graded C for recursive force delete. lifeos is graded C for downloading remote code and making network calls. plan-devex-review carries a low-confidence network finding while remaining A.
The grade summarises what the scan treated as risky. Each finding shows the file line that prompted it, so you can read the evidence and decide whether it fits your use.
Six planning paths for different kinds of work
The original is the first copy of a file we saw; everything matching it afterwards is a copy. Every pick below is original.
- brainstorming: turns an idea into a design or implementation plan before code changes. 37 tokens every session, 251 lines, 13 copies.
- spec_generator: turns an issue and repository findings into a structured work specification in JSON. 17 tokens every session, 99 lines.
- plan-devex-review: reviews developer plans, APIs, tools, libraries, and platforms for confusing steps. 15 tokens every session, 1,146 lines.
- aip-user-stories: turns Apache Airflow Improvement Proposals into user-story playbooks. 56 tokens every session, 210 lines.
- sparc-methodology: guides work through specification, pseudocode, architecture, refinement, and completion. 29 tokens every session, 1,116 lines, 9 copies.
- constraint-driven-development: records measurable quality rules and checks that implementation does not weaken them. 176 tokens every session, 312 lines, 1 copy.
When a rule or hook fits better
Choose a rule when the instruction belongs to a file type rather than a whole planning process. A glob is a filename pattern, like *.tsx. The rule can then wait for matching files instead of entering planning work that does not concern them.
Choose a hook when an action must run after an event. A hook is a shell command your agent runs for you when something happens. A skill cannot fire on an event, although it may tell the model to run a script.
Questions people ask
Which planning skill should I choose for a vague feature? Choose brainstorming when you cannot yet explain the intended result or agree on the requirements. Move to a more specific workflow after that uncertainty is settled.
Will a planning skill cost me if I never call it? Almost always. Only 655 of the 26,049 wait to be requested; the other 25,394 put their name and description in every session whether you use them or not.
When should I use a shared planning map? Choose wayfinder when the work needs decisions handed between people or sessions. It has been copied into 34 other projects, which is as close as the catalogue comes to a vote.
Do it now
Run npx agentmods cost to see what your project already pays, then browse the skills catalogue.