skills: Instructions file for Claude Code

AGENTS.md

Guidance for AI agents (and humans) working in this repository. The history and evidence behind its rules live in docs/decisions.md.

What this repo is

Public Agent Skills for Scenario: procedural knowledge that teaches AI coding agents to create production-ready content through the Scenario MCP server. The content spans images, video, audio, textures, skyboxes, 3D, and custom models, for games, entertainment, and any creative vertical.

• Format: Agent Skills.
• Install: npx skills add scenario-labs/skills.
• Listing: skills.sh.
• The one exception to the MCP focus: the expert tools, teams of skills that drive DCC software and game engines installed on the user's machine (see Expert tools).

Quick start

• pnpm install: once after cloning. It installs the git hooks.

• pnpm format && pnpm validate: before committing. format rewrites files locally; validate runs the checks CI runs, format:check included.

• pnpm test: after touching a script. Through uv, each Python suite gets the packages its tests/<name>/requirements.txt lists.

• /skills:validate <name> ($skills-validate in Codex): the application test, before merging a skill change.

Layout

skills/<name>/SKILL.md                         # name must equal the directory name
skills/<category>/<family>/README.md           # expert tools: the family's build notes
skills/<category>/<family>/<name>/SKILL.md     # expert tools: dcc/ or game-engines/
tests/<name>/                                  # test suite for any script shipped with skill <name>

• scripts/lib/skills.mjs defines this layout, and every check finds skills through it.
• Skill names are unique repo-wide (pnpm skill-files checks). The skills CLI installs every skill flat under its name, whatever its folder, and silently keeps only the first of two that share one.
• A supporting file (heavy reference, script) sits next to a SKILL.md only when its content is too large to inline.
• Link every supporting file directly from SKILL.md. Agents resolve file references one level deep, so a file reached only through another may never be read.
• A skill may carry a README.md recording how it was built and why it makes its choices. It is for the agents and humans working on the skill, not the agent running it.
  • It is the one supporting file exempt from the link rule (pnpm skill-files skips it).
  • Leave it unlinked: linking it spends body words and invites a runtime agent to read maintainer notes as instructions, which once made an agent discount a fact it needed.
• Every script shipped with a skill has a test suite in tests/<name>/ at the repo root, never in the skill directory: a published skill directory carries only what an agent needs at runtime. See Validation and testing.