Skills

A skill is a reusable workflow playbook the agent can apply to a task. Skills also anchor the distillation flywheel: high-rated interactions on a skill are collected, curated, and exported as a training dataset — the seed for sharpening that skill over time.

Working with skills

GET/v1/skills —

list skills.

GET/v1/skills/{slug} —

fetch one skill.

POST/v1/skills —

create a skill. Fields: slug (required), name (required), description, base_model_id.

GET/v1/skills/_candidates —

surfaces recurring task shapes that are good candidates to become skills.

bash

curl https://nexevo.ai/v1/skills \
  -H "Authorization: Bearer $NEXEVO_API_KEY"

The distillation flywheel

Each skill accumulates a training corpus from the interactions where it did well:

POST/v1/skills/{slug}/collect —

gather the eligible (highly-rated) samples for the skill.

POST/v1/skills/{slug}/curate —

keep / drop samples to clean the set.

POST/v1/skills/{slug}/export —

export the curated set as an OpenAI-chat JSONL dataset (the seed for fine-tuning that skill).

At run time the agent pulls a relevant skill's playbook on its own (it does not have to be named in the request). Creating sharp, well-described skills makes the agent more consistent on your recurring workflows; the collect → curate → export loop is how a skill's quality compounds with use.