The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests and processes search results from external, untrusted sources like npm, PyPI, and GitHub. This creates a surface for indirect prompt injection where malicious package descriptions or metadata could attempt to influence the agent's evaluation or subsequent actions.
Ingestion points: External search results from public registries and web searches (SKILL.md).
Boundary markers: Absent. The prompt template for the research agent does not include specific delimiters or warnings to ignore instructions embedded in the search results.
Capability inventory: The skill encourages package installation (npm, pip) and the execution of discovered tools (SKILL.md).
Sanitization: No explicit sanitization, validation, or filtering of the external content is described.
[EXTERNAL_DOWNLOADS]: The skill's core functionality involves identifying and recommending the installation of third-party packages from well-known registries such as npm and PyPI. While these are established services, the workflow relies on the integrity of these external dependencies.
[COMMAND_EXECUTION]: The workflow involves executing commands to search the local environment (e.g., rg) and installing discovered packages (e.g., npm install). The skill also accesses local configuration files such as ~/.claude/settings.json to identify active MCP servers.

search-first