The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes financial documents and filings sourced from the SEC EDGAR database. These documents are external, untrusted data that could contain malicious instructions or patterns designed to influence the behavior of the AI agent when the content is converted to Markdown or included in the context via .to_context().
Ingestion points: Found in SKILL.md and references/filings.md via methods like filing.text(), filing.markdown(), and filing.to_context().
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' commands when processing this external content.
Capability inventory: The skill utilizes network operations to fetch data and includes a filing.open() command that can trigger browser actions.
Sanitization: No sanitization or validation of the text content extracted from SEC filings is described.
[PERSISTENCE]: The skill provides automated installation methods and manual configuration steps to maintain its presence in the agent's environment.
Evidence: In references/ai-integration.md, the install_skill() function writes data to ~/.claude/skills/edgartools/, and the MCP setup instructions involve modifying persistent configuration files like claude_desktop_config.json.
[METADATA_POISONING]: The skill metadata contains inconsistent authorship information. The SKILL.md file identifies the author as K-Dense Inc., which differs from the system record of leonchaox. This discrepancy could be used to misrepresent the provenance of the skill.

edgartools