The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill includes a dedicated 'Credential Safety' section in workflow.md that instructs the agent to avoid committing secrets, authentication tokens, or .env files. It correctly identifies sensitive file paths (e.g., ~/.npmrc, .env) and advises using environment variables instead of hardcoded secrets, which is a security best practice.
[EXTERNAL_DOWNLOADS]: The skill references external resources and installation commands for related skills (e.g., npx skills add ariestools/ariestools-skills). These references target the official GitHub organization and repositories belonging to the skill's author ('ariestools') and are consistent with the intended functionality of the platform's skill management system.
[COMMAND_EXECUTION]: The skill provides instructions for executing standard development commands using git, pnpm, npm, and yarn. These commands are limited to routine tasks such as building, linting, testing, and dependency management within a local repository context. It explicitly warns against running ad-hoc commands that bypass the project's defined toolchain.
[INDIRECT_PROMPT_INJECTION]: The skill involves reading and processing untrusted data from repository files like AGENTS.md, CLAUDE.md, and PRD.md. It mitigates this risk by establishing a clear hierarchy of authority (repository-specific instructions override defaults) and defining a robust 'Definition of Done' that requires verification through builds, lints, and tests before a task is considered complete.

xy-development