The Agent Skills Directory

REMOTE_CODE_EXECUTION (HIGH): The skill instructions require the user to execute npx -y octagon-mcp, which downloads and runs code from the npm registry. The author OctagonAI is not a trusted provider, and the execution of unverified remote code is a critical security risk.
COMMAND_EXECUTION (HIGH): The setup process for both Windows and macOS involves manual shell command execution to set environment variables and run remote packages (e.g., cmd /c "set OCTAGON_API_KEY=...").
EXTERNAL_DOWNLOADS (MEDIUM): The documentation suggests installing Homebrew using a curl | bash pipe from raw.githubusercontent.com. While the domain is whitelisted for exfiltration, the specific repository (Homebrew/install) is not on the trusted list, and piping remote scripts directly to a shell is a dangerous practice.
PROMPT_INJECTION (HIGH): The skill is highly vulnerable to indirect prompt injection (Category 8). It is designed to ingest and analyze external earnings call transcripts, which are untrusted data sources.
Ingestion points: Earnings call transcripts are processed via octagon-transcripts-agent in multiple workflow phases.
Boundary markers: There are no delimiters or instructions to ignore embedded commands within the processed transcripts.
Capability inventory: The skill utilizes tools with high capability, including octagon-scraper-agent (web scraping), octagon-deep-research-agent, and octagon-agent (general market intelligence).
Sanitization: No evidence of sanitization, filtering, or validation of external content before it is interpolated into agent prompts.

earnings-analyst-master