The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions explicitly permit the agent to use the shell to execute commands when native platform tools are unavailable (e.g., using bundle show). This grants the agent the ability to perform arbitrary command execution on the host environment.
[DATA_EXFILTRATION]: The research methodology directs the agent to recursively search for skill files in user home directories, specifically paths like ~/.claude/skills/, ~/.codex/skills/, and ~/.agents/skills/. This facilitates traversal and data access outside of the immediate project workspace.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from untrusted external sources, creating an attack surface for indirect prompt injection.
Ingestion points: Retrieval of SKILL.md files from multiple filesystem locations and extraction of external content from websites and documentation via the Context7 MCP tool.
Boundary markers: The instructions lack specific delimiters or instructions to ignore commands or instructions embedded within the researched data.
Capability inventory: The agent possesses capabilities to execute shell commands, read local files, and perform network-based web searches.
Sanitization: No procedures are defined for sanitizing, escaping, or validating the content retrieved from external sources before it is interpreted by the agent or delivered as actionable guidance.

best-practices-researcher