The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by reading card fields via %{field:<id>} tokens and pulling context from knowledge base items (plain text, documents, and data lookups).
Ingestion points: Card field identifiers and knowledge base data sources described in SKILL.md.
Boundary markers: There are no instructions for the agent to use specific delimiters or to ignore instructions embedded within the ingested data.
Capability inventory: The agent has extensive capabilities including updating card data, moving cards between phases, sending emails, and executing other MCP tools.
Sanitization: The skill description does not include procedures for sanitizing or validating ingested data before it is used in prompts.
[DATA_EXFILTRATION]: The skill allows the agent to interact with local system file paths through the file_path parameter in the create_ai_knowledge_base_document tool and the configuration_file_path in provider configuration tools. While intended for document indexing and provider setup, this parameterization could be exploited to access sensitive local files if the agent is directed to unintended paths.
[DYNAMIC_EXECUTION]: The mcp_tool action type allows for the dynamic invocation of other MCP tools by name, enabling the agent to chain multiple tool behaviors at runtime based on its instructions.
[COMMAND_EXECUTION]: The skill includes a comprehensive CLI reference (references/cli.md) that maps agent management tasks to shell commands, facilitating direct command execution in a terminal environment.

pipefy-ai-agents