The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external workflow files (BPMN, Langflow JSON, n8n JSON) that may contain untrusted data provided by the user. This creates a surface for indirect prompt injection where malicious instructions could be embedded in the source files.\n- Ingestion points: The skill ingests content from external files via the @ reference syntax, specifically extracting data from flow metadata, node templates, code fields, and global variables.\n- Boundary markers: The instructions include strict 'Documentation Principles' and 'Quality Standards' that mandate staying factual and avoiding speculation or assumptions, which provides a process boundary to mitigate accidental obedience to embedded instructions.\n- Capability inventory: The skill instructions are limited to text analysis and natural language generation; no dangerous tools (such as network access, system command execution, or file writing) are requested or utilized.\n- Sanitization: The skill implements a translation layer (converting technical data into business summaries), though it explicitly requires the preservation of LLM prompt text and code logic from the source files in sections 6 and 6.1.

sop-builder