The Agent Skills Directory

REMOTE_CODE_EXECUTION (CRITICAL): The references/mcp-setup.md file contains an instruction to install Homebrew using the pattern /bin/bash -c "$(curl -fsSL ...)". Piping remote scripts directly into a shell is a high-risk pattern that bypasses local security verification.
EXTERNAL_DOWNLOADS (HIGH): The skill requires the installation of the octagon-mcp package via npx. The 'OctagonAI' organization is not on the trusted sources list, meaning the code being executed at runtime is unverified and could be modified by the package owner without notice.
COMMAND_EXECUTION (MEDIUM): Installation and configuration require the execution of shell commands (env, set, npx) that handle sensitive information like the OCTAGON_API_KEY. This exposes the credential to the local process environment and command history.
PROMPT_INJECTION (MEDIUM): As an indirect injection risk, the skill utilizes tools (octagon-agent, octagon-scraper-agent) that ingest untrusted data from SEC filings and financial websites.
Ingestion points: Market data, SEC filings, and external website content via octagon-scraper-agent.
Boundary markers: None detected in the prompt templates (e.g., <prompt> or {{prompt}} are used without delimiters).
Capability inventory: The agent can process this data to provide financial analysis and investment signals, influencing the user's decision-making process.
Sanitization: No evidence of sanitization or filtering for the external content before it is processed by the AI.

sector-performance-snapshot