The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the installation of plugins from arbitrary, unverified GitHub repositories discovered through a public topic search. The installation process (e.g., pnpm dsh plugin add) executes remote code via package lifecycle scripts (preinstall, postinstall) from these unknown sources.
[CREDENTIALS_UNSAFE]: The included search script scripts/search-topic.mjs retrieves the user's GitHub authentication token by executing gh auth token to increase API rate limits. This constitutes access to sensitive local credentials.
[COMMAND_EXECUTION]: The skill executes various shell commands for plugin discovery and system configuration, including running internal scripts and package managers.
[PROMPT_INJECTION]: The skill processes untrusted content (READMEs and package manifests) from remote GitHub repositories to determine installation methods, creating a surface for indirect prompt injection.
Ingestion points: GitHub repository metadata, README.md files, and package.json manifests from discovered plugins.
Boundary markers: None identified.
Capability inventory: Shell command execution (pnpm, node) and file system operations.
Sanitization: The instructions only ask the agent to manually look for 'extra downloads' in READMEs, which is not a robust security control.

find-plugins