Originals and copies: find the first skill

15,698 skills measured: learn to spot the original, read copy counts, compare token preload, and choose where a copied file deserves your trust.

You find the same useful instructions in two repositories and are about to install the second one. Copied files keep their old text when the first file changes. The catalogue shows which file came first before you decide.

Find the first file before you trust the second

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 original is the first copy of a file we saw; everything matching it afterwards is a copy. A copy is a file whose body matches another file we saw earlier; the earlier one is called the original.

“Original” is a catalogue label, not a claim about who invented the idea. One entry already has 46 copies, vercel-composition-patterns. If its original changes, those copies do not follow automatically.

The preload is small, but it repeats

A token is the unit your AI plan bills by, about three quarters of a word. Skills loaded every session have a median preload of 54 tokens. Nine in ten stay under 132.

The largest every-session preload is 747 tokens in ai-pm-prompts. At 20 sessions a day, the 54-token median becomes 1,080 tokens across a day. That is close to 1,296 tokens, the catalogue median for one CLAUDE.md file.

405 skills wait to be asked for, so they do not add their full body before the model chooses them. A copy can cost the same as its original while giving you older instructions.

The scan finds capabilities, not a verdict

The scan grades 14,844 skills A, 466 B, 388 C or D, and 0 E or F. It flagged 1,496 of them for network calls, 194 for asking for root access and 151 for downloading and running remote code.

Another 134 were flagged for shell commands, 126 for a recursive force delete and 111 for reading agent configuration directories. The core, timeline-report, and scrapling-official entries each carry a network-call finding.

A finding gives you a line to read. It describes what the scan saw, not whether you should install the mod. The page shows the file line that produced each finding.

Six original skills with visible copy trails

The figures after each link show preload tokens, file length, original status, and known copies.

  • ppt-generation: creates PowerPoint presentations by designing slides and assembling a PPTX file. 44 tokens every session, 464 lines, original, 7 copies.
  • vercel-composition-patterns: gives React patterns for flexible, reusable interface components. 58 tokens every session, 89 lines, original, 46 copies.
  • context-restore: restores saved work so you can continue after changing sessions or branches. 18 tokens every session, 562 lines, original, 1 copy.
  • core: guides browser control, page inspection, screenshots, forms, and sign-ins. 112 tokens every session, 541 lines, original, 3 copies.
  • hyperframes: creates timed HTML video compositions with captions, audio, and motion. 155 tokens every session, 503 lines, original, 4 copies.
  • cpp-coding-standards: applies modern C++ guidance when writing, reviewing, or refactoring code. 48 tokens every session, 724 lines, original, 14 copies.

Choose a hook when the task must fire on an event

A hook is a shell command your agent runs for you when something happens. Choose one when the job must happen before a tool call, after a write, or at another event.

A skill cannot fire on an event. It waits for the model to decide it is relevant. Use a skill for guidance, and a hook when timing must control an action.

Questions people ask

How do I tell if a skill is copied? Check the original or copy label and the copy count. If two entries match, start with the original unless the other repository is the one you trust.

Do copies stay current? No. When the original changes, a copy keeps its old text, and 5,575 of these files changed in the last 7 days. Choose a copy only when its local version is deliberately different.

Will every skill add tokens before I ask? No. 15,293 skills load every session, while 405 wait to be asked. Those are counts of skills, not token costs.

Do it now

Run npx agentmods add <slug> to install a chosen entry, then browse the skills catalogue.


Every figure here comes from the catalogue at agentmods.dev, which re-crawls itself nightly. More writing · how the data is licensed.