Skip to content

Glossary

onfire7777 edited this page Jun 29, 2026 · 1 revision

Glossary

Key terms used throughout the Universal AI Skills Library and this wiki.

Term Definition
UASL Universal AI Skills Library — this project.
Router-first The architecture principle that clients reach skills through a router, never by holding their own copy. See Architecture.
skill-router The single Go CLI binary that routes prompts to skills, loads them, generates the registry, and serves MCP. See Skill Router CLI.
Corpus The set of all skills, stored once under skills/. See Skills Corpus.
Skill A directory skills/<id>/ containing a SKILL.md (+ optional files) that teaches an agent how to do something.
SKILL.md The required file in every skill: YAML frontmatter + a markdown body.
Registry / Manifest manifest.json — the generated, machine-readable index of the corpus and the only contract between corpus and router. See Manifest & Registry.
Canonical ID A skill's kebab-case top-level directory name; the authoritative identifier.
Alias A legacy or display name that resolves to a canonical ID via compatibility_aliases.json.
Preflight The deterministic decision step that returns route / ambiguous / no_route for a prompt, with no extra LLM call.
Route A preflight decision that one skill is a confident match (also the name of the route verb/command).
Ambiguous A preflight decision with several candidate skills; the host AI chooses.
No-route A preflight decision that no skill is relevant; the agent proceeds normally.
Compose / Pipeline Building a multi-step plan (DAG) that chains skills — skill-router compose --pipeline.
skillservice The internal Go engine both the CLI and the MCP server call — the single routing seam.
MCP Model Context Protocol — the tool-server interface exposed by skill-router serve. See MCP Server.
Wrapper / Adapter The compact, per-client instruction or skill that calls the router instead of embedding the corpus.
Wrapper install The default install mode: a tiny wrapper, not the whole corpus. See Installation & Setup.
Universal AI Stack The optional local runtime (ai-setup/) that routes model calls across providers with health checks and a guarded local fallback.
Byte-parity The guarantee that the Go registry generator emits output byte-for-byte identical to the legacy Node generator (make parity). See Node → Go Migration.
Routing-eval The harness (tests/routing-eval/) that measures routing quality with P@1, MRR, Recall@5, and abstention accuracy. See Testing & CI.
P@1 Precision at rank 1 — how often the top-ranked skill is correct.
MRR Mean Reciprocal Rank — average of 1/(rank of the first correct skill).
Recall@5 Whether a correct skill appears within the top 5 results.
Abstention accuracy How reliably the router refuses to route on off-domain prompts.
Characterization test A test that pins existing behavior so refactors don't change it (tests/characterization/).
Public-safe Defaults that keep machine state and secrets out of the repo, enforced by audits and gitleaks.

See also: FAQ · Architecture · Home.

Clone this wiki locally