The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): The installation instructions in README.md and references/mcp-setup.md direct users to execute remote code via npx -y octagon-mcp and npx skills add OctagonAI/skills. These packages/scripts originate from 'OctagonAI', which is not a trusted source according to security policy. This allows for arbitrary code execution on the user's machine during setup.
[REMOTE_CODE_EXECUTION] (MEDIUM): references/mcp-setup.md instructs users to install Homebrew using a 'curl | bash' pattern (/bin/bash -c "$(curl -fsSL ...)"). While Homebrew is a common tool, this pattern is a classic RCE vector and the source organization is not explicitly on the trusted whitelist.
[PROMPT_INJECTION] (MEDIUM): The skill is susceptible to Indirect Prompt Injection (Category 8) as it is designed to ingest and process external content from SEC EDGAR filings.
Ingestion Points: Item 1 (Business Description), Item 1A (Risk Factors), Item 7 (MD&A), and Footnotes from 10-K, 10-Q, and 8-K filings.
Boundary Markers: Absent. There are no instructions or delimiters provided to the agent to treat filing text as untrusted data or to ignore embedded instructions.
Capability Inventory: The skill generates long-form institutional-quality research reports (up to 15,000 words), meaning malicious content in a filing could significantly bias the agent's analysis and output.
Sanitization: Absent. No evidence of escaping or filtering logic for the ingested external text.
[EXTERNAL_DOWNLOADS] (MEDIUM): The skill documentation encourages downloading tools and configuration from github.com/OctagonAI, which is not a verified/trusted repository.

sec-analyst-master