Skip to main content
PRYSM ships 23 production-grade engineering skills (sourced from addyosmani/agent-skills, MIT) — structured expert instructions for code review, debugging, TDD, documentation, and more. Skills act as expert system prompts: invoking a skill routes the task to the best model for that intent and returns a signed completion.

List skills

GET https://api.prysm1.com/v1/skills · No authentication required
Returns all available skills with their IDs, names, and descriptions.

Get skill

GET https://api.prysm1.com/v1/skills/{skill_id} · No authentication required
Returns a skill’s full content: parsed frontmatter (name, description) and the raw markdown body that becomes the system prompt when the skill is invoked.

Invoke skill

POST https://api.prysm1.com/v1/skills/{skill_id}/invoke · Requires Authorization: Bearer prysm_sk_…
Run a skill on a task. The skill’s full body becomes the system prompt; your task is the user message. PRYSM routes to the best model for the skill’s intent — a code-review skill routes to a premium reasoning model; a summarisation skill to a fast budget model. All standard PRYSM features apply: BRAIN.md overrides, AgentGuard cost caps, and a PrysmProof receipt. The response includes prysm.skill — the invoked skill’s identity — alongside the usual prysm.routing, prysm.cost, and prysm.proof fields.

Request body

string
required
The user’s prompt — what you want the skill to do.
string
default:"auto"
Model override. auto lets PRYSM pick; pass laser-1, halo-1, or foton-1 to fix the intensity tier.
string
Explicit routing mode: quality, balanced, or agility.
integer
Maximum output tokens. Defaults to the model’s limit.