The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to run npx @claude-flow/cli@latest to search and store patterns. Executing unpinned packages from external repositories at runtime creates a vulnerability to supply chain attacks or package hijacking.
[INDIRECT_PROMPT_INJECTION]: The skill presents an injection surface through data ingestion pipelines.
Ingestion points: Loads external OHLCV market data via mcp__plugin_ruflo-core_ruflo__memory_search from the market-data namespace in SKILL.md.
Boundary markers: Absent. There are no boundary delimiters or warnings instructing the agent to ignore prompt instructions embedded in the market data.
Capability inventory: The skill has access to the powerful Bash tool as defined in the allowed-tools metadata field.
Sanitization: Absent. The data is parsed and processed directly without validation or sanitization routines.

market-pattern