Skip to content

uniac-ai/uniac

v0.3.12MIT

Deploy and run applications on Uniac, the cloud deployment platform — the uniac.yaml manifest, the uniac CLI, and the platform contract.

GitHubOpen Website

agent-skills

The public agent knowledge for Uniac, a cloud deployment platform. Everything an agent needs to build on Uniac ships from this repository: the uniac skill (with its bundled contract references), and the agents.md bootstrap document the website serves to visiting agents.

The audience is a consumer of Uniac — an agent (or the engineer directing it) with an application that needs to be running. Nothing here assumes access to Uniac's own source, and nothing here documents how the platform is built.

Install

npx skills add uniac-ai/agent-skills -g

Installs the uniac skill for the coding agents on the machine. Re-run the same command to update. Once the discovery index is served, the domain form works too: npx skills add uniac.ai.

This is the one install line to publish: the ecosystem leaderboard counts installs per repository slug, so every mention should aggregate under the same one.

In Claude Code the repository is also a plugin marketplace:

/plugin marketplace add uniac-ai/agent-skills
/plugin install uniac@uniac

/plugin marketplace update uniac pulls later releases.

Layout

skills/uniac/                 the one skill
  SKILL.md                    resident knowledge: mental model, loop,
                              commands, environment, sharp edges
  references/manifest.md      full uniac.yaml schema, reference grammar,
                              plan-time rules, multi-service composition
  references/cli.md           what each command does and needs, the output
                              contract, typed exit codes, headless env vars
agents/agents.md              the bootstrap page uniac.ai serves to agents —
                              machine setup and sign-in, before knowledge
                              is installed
tools/validate.py             frontmatter and link gate (what strict
                              installers reject, CI rejects first)
tools/generate_manifests.py   the plugin name, release, blurb, licence and
                              links, and everything rendered from them
.claude-plugin/               Claude Code marketplace + plugin manifest
.codex-plugin/plugin.json     Codex plugin manifest
plugin.json                   agent-plugins.org manifest (Cursor imports it)
.well-known/agent-skills/     the discovery index and the skill archive
                              it points at
LICENSE                       MIT — required by the Cursor marketplace

Every file under those last four entries is generated. tools/generate_manifests.py holds the facts they share; CI runs it with --check, so editing one of them by hand fails the build.

One skill by design: an agent's Uniac task always needs the mental model, the manifest, and the CLI together, so they load as one body with the deep contracts as on-demand references. A second skill appears only when a distinct activity earns it.

The boundary with agents/agents.md: the bootstrap document carries only what an agent needs before the skill is installed (consent etiquette, the install line, sign-in); the skill carries everything after. Neither restates the other.

Publishing

  • Every claim is verified against the released uniac binary — prefer having run the command over having read about it. A wrong field is worse than a missing one; the skill is read by agents that cannot check it.
  • Carry what is durable and unobservable; drop what churns or can simply be run. The reader is a coding agent with a shell, so the skill says what a command does and what the surface is, and leaves the rest to be looked at. The test for a sentence: if a change to the CLI's rendering would falsify it, it is presentation — cut it, and name the command that shows it. Three things stay however small they look: hazards that must not be discovered by experiment, what has to be known before anything can run (the manifest schema), and statements about output that survive a re-render.
  • The verification stamp is VERSION in tools/generate_manifests.py. It is the release the contracts were checked against, and the version every plugin manifest carries.
  • At each CLI release the contracts are re-verified against, bump that constant, regenerate, and tag this repository v<cli-version>. The UniacWeb build vendors the tag matching the released CLI and serves agents.md, the skill files, and the discovery index same-origin at uniac.ai.
  • Before pushing: python3 tools/validate.py and python3 tools/generate_manifests.py; CI runs both plus a live resolve through the ecosystem installer (npx skills add . --list).