The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides a template for a 'calculatorTool' that uses the unsafe eval() function to process mathematical expressions derived from user input.
Evidence: const result = eval(expression); // Use safer math parser in production within the tools/custom.ts snippet in SKILL.md.
The use of eval() on strings that may be influenced by user input or LLM generation creates a significant remote code execution risk, despite the inclusion of a warning comment.
[INDIRECT_PROMPT_INJECTION]: The skill implements complex workflows that ingest untrusted external data, creating an attack surface for indirect prompt injection.
Ingestion points: The skill uses DirectoryLoader to ingest PDF, text, and JSON files, and PineconeStore to retrieve context for RAG pipelines.
Boundary markers: The RAG prompt template (rag/chain.ts) lacks clear delimiters or 'ignore' instructions for the {context} variable, which is populated from external documents.
Capability inventory: The agents constructed using these patterns have access to the calculatorTool (which executes eval) and a dbQueryTool capable of performing database lookups.
Sanitization: The examples do not demonstrate sanitization or filtering of retrieved document content before it is interpolated into the system prompt.
[COMMAND_EXECUTION]: The skill documentation instructs users to execute shell commands to install dependencies.
Evidence: npm install langchain @langchain/openai @langchain/community in the 'Installation' section of SKILL.md.
While these are standard package installations, they involve local command execution as part of the setup process.

langchain-workflow-builder