Imported from linkmodelhq/linkmodel-skills (AGENTS.md). Install upstream with npx skills add linkmodelhq/linkmodel-skills. Copyright stays with the author.

Repository Guidelines

Project Structure & Module Organization

This repository publishes reusable LinkModel agent skills. Each skill lives under skills/<skill-name>/ and keeps its entry point in SKILL.md. Put executable helpers in scripts/, agent metadata in agents/, prompts and catalogs in references/, and behavior tests in the skill's tests/ directory. Repository-level validation lives in tests/. Design specifications and implementation plans belong in docs/superpowers/; update README.md and skills.sh.json when adding or renaming a public skill.

Build, Test, and Development Commands

There is no compilation step or package-local dependency installation.

bash tests/validate-repository.sh

Runs shell syntax checks, the mocked end-to-end Skill tests, repository structure checks, and the English-only scan for skills/.

./skills/generate-linkmodel-brand-visuals/scripts/generate-visual.sh \
  --brief "A dramatic product scene" --dry-run

Previews prompt assembly and commands without making a billable request.

Coding Style & Naming Conventions

Write portable Bash with #!/bin/bash, set -eu, two-space indentation, quoted expansions, and explicit input validation. Use snake_case for shell variables, UPPER_SNAKE_CASE for exported constants, and kebab-case for Skill directories, preset IDs, and script filenames. All source code, comments, Skill content, command output, tests, contributor documentation, and commit messages must be English. Preserve tab-delimited formatting in catalog files.

Testing Guidelines

Add regression coverage to the affected Skill's tests/test-skill.sh. Use isolated mktemp workspaces, mock network-facing CLIs, and verify successful artifacts plus expected failures. Tests must not require API keys, overwrite production assets, or make billable calls.

Commit & Pull Request Guidelines

Use scoped Conventional Commits, for example feat(visuals): add custom brand visual generation. Accepted prefixes are feat, fix, refactor, chore, and docs.

Pull requests should explain the affected Skill, user-visible behavior, and validation performed. Link relevant issues or design documents, include sample output for visual changes, and identify external CLI, configuration, or deployment requirements.