Four steps,every time your AI asks.
Discover, assess, match, load. The result tells your AI what to do next.
Discover
Edge keeps its own list of over 100,000 public skills, with what each one says it does.
→Assess
It considers available provider reports; Edge checks the exact bundle before load. Skills that fail the security gate are withheld when Edge has scan evidence.
→Match
Your AI describes the task. Edge finds skills by meaning, then an AI ranks a shortlist for the task when ranking is available.
→Load
Your AI gets the skill’s instructions for this task only. Nothing is installed.
A closer look
One card per step. Technical details stay folded.
Where the skills come from
- Public skills listed on skills.sh, written by the community and by companies.
- The current served index holds over 100,000 public skills. The count changes as the index updates.
- Edge also publishes a few dozen skills of its own.
What “checked” means
- Edge reads available skills.sh provider reports and runs its own checks on the exact bundle before load.
- Two failing scanners, or a high or critical finding, cause Edge to withhold a skill when it has that evidence.
- A skill without enough current scan evidence can appear with a security note; loading triggers an exact-bundle check and can be withheld.
How it picks
- Search by meaning plus keywords finds 20 candidates.
- An AI ranker judges each one against your exact task and keeps the best five. Free ranking currently has no enforced daily cap. If ranking is unavailable, Edge keeps catalog order.
- If nothing fits, Edge says so. “None” is a valid answer.
What your AI receives
- The skill’s instructions as text, with a note to read them before running anything.
- For this task only. Nothing lands on your disk.
- Edge never runs the skill. Your AI decides what to do with it.
For developers: discovery, screening and loading
find_skill retrieves a candidate pool from Edge’s own index and may rank it for the task. It considers available provider reports and can show a scan-pending note. use_skill loads one result only after its exact pinned bundle satisfies Edge’s current security gate. Ratings are recorded but do not change ranking in the current release.
See exact tool inputs and audit states.