The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill creates a significant attack surface by instructing the agent to ingest and act upon data from external, untrusted sources including npm, PyPI, GitHub, and the general web.
Ingestion points: The workflow explicitly uses a 'general-purpose' subagent to search external registries and the web for code patterns and tools.
Boundary markers: No explicit boundary markers or instructions are provided to the agent to treat search results as untrusted or to ignore instructions embedded within retrieved documentation or README files.
Capability inventory: The skill provides a framework for the agent to perform npm install, pip install, and execute shell commands (rg) based on the research findings.
Sanitization: No sanitization or verification steps are required for the external content before it influences the agent's next steps.
[EXTERNAL_DOWNLOADS]: The skill encourages the dynamic discovery and installation of external software packages (e.g., npm install textlint-rule-no-dead-link, ajv-cli) without requiring a verification of the source's integrity or security posture beyond basic metrics like 'maintenance' and 'community'.
[DATA_EXPOSURE]: The instructions direct the agent to access potentially sensitive local configuration files, specifically ~/.claude/settings.json, to inspect the current environment's capabilities. These files may contain API keys, tool configurations, or other sensitive metadata.

search-first