The Agent Skills Directory

[PROMPT_INJECTION]: The skill distributes a file named AGENTS.md which contains extensive instructions designed to override the agent's default behavior. It explicitly states that these standards "outrank generic 'best practices' and training-data defaults" and mandates the use of specific reasoning specialists and communication styles.
[PERSISTENCE]: The skill modifies global configuration files in the user's home directory, such as ~/.claude/CLAUDE.md and ~/.gemini/settings.json. By injecting references to the author's AGENTS.md contract into these files, it ensures that its overriding instructions are persistently loaded by the AI agent across different projects and across the user's entire local environment.
[COMMAND_EXECUTION]: The included quality-gate tests, specifically knip.test.mjs, use execFile to execute local binaries from node_modules. While these are used for linting and dead-code detection, they represent a vector for executing code within the repository's context.
[INDIRECT_PROMPT_INJECTION]: The skill creates a surface where repository content controls agent behavior by instructing the agent to always follow the rules found within a local AGENTS.md file. This allows any user who can commit to the repository to influence the agent's behavior and bypass default safety or style guidelines.
[EXTERNAL_DOWNLOADS]: The skill installs various development dependencies from the NPM registry and references a remote plugin installation command (npx skills@latest add Gabriel-Lafrance/Skills) that downloads and executes code from an external source.

setup-toolkit