The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to locate and access sensitive credential files (.env) across the filesystem, including parent directories, which can lead to unauthorized data exposure.\n
Evidence in SKILL.md: A setup script is provided that searches for API keys by walking up the directory tree from the current working directory (Path.cwd().resolve().parents).\n
Evidence in SKILL.md: The script specifically probes for files named .env, document-extraction/.env, and skills/document-extraction/.env at each level of the directory hierarchy.\n
Evidence in SKILL.md: The documentation explicitly directs the agent to verify the existence of sensitive files using system commands: ls -la .*/skills/document-extraction/.env.\n\n- [INDIRECT_PROMPT_INJECTION]: The skill serves as a gateway for external, untrusted content to enter the agent's execution context, creating a vulnerability to indirect prompt injection attacks.\n
Ingestion points: External document data is ingested via the document (local file/bytes) and document_url (remote URL) parameters in LandingAIADE.parse(), as well as the markdown_url parameter in the extract() and split() methods documented in SKILL.md and references/use-cases.md.\n
Boundary markers: There are no explicit instructions or mechanisms provided to wrap the extracted markdown content in protective delimiters or to warn the agent against executing instructions embedded within the documents.\n
Capability inventory: The skill provides extensive document parsing, hierarchical classification, and structured data extraction capabilities in SKILL.md. The resulting markdown content is intended to be used by the agent to drive further actions, such as writing code or summarizing information.\n
Sanitization: The skill does not implement any sanitization, filtering, or validation of the extracted document content to neutralize potential malicious prompts hidden in the source files.\n\n- [DYNAMIC_EXECUTION]: The documentation includes instructions for executing dynamically generated scripts in the local environment.\n
Evidence in SKILL.md: The "API Key Setup" section provides a shell command that uses a heredoc (`python
<< 'EOF'`) to execute a Python script constructed at runtime to perform the filesystem search for credentials.

document-extraction