The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents a wide range of pnpm CLI commands for package management, environment configuration, and workspace operations. This includes commands that execute shell scripts such as pnpm run, pnpm exec, and pnpm dlx.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and act upon configuration files which may contain untrusted data or instructions.
Ingestion points: Files such as package.json, pnpm-workspace.yaml, .npmrc, and .pnpmfile.cjs are identified as primary sources of configuration and scripts (found in references/core-config.md and references/features-hooks.md).
Capability inventory: The agent is instructed to run scripts and execute binaries defined within these project files, which can lead to arbitrary command execution.
Boundary markers: The skill does not provide specific instructions or markers for the agent to distinguish between administrative instructions and untrusted content within these project files.
Sanitization: There are no explicit sanitization steps mentioned for processing the contents of these configuration files before execution.
[DYNAMIC_EXECUTION]: The skill provides detailed guidance on using .pnpmfile.cjs hooks. These hooks allow users to write and execute arbitrary JavaScript logic that can programmatically alter package resolution and dependency metadata at runtime (found in references/features-hooks.md).

pnpm