The Agent Skills Directory

Prompt Instruction Override: The skill uses directives such as "CRITICAL" and tells the agent its "knowledge of workflow is outdated." These patterns are designed to override the agent's internal training data in favor of the instructions provided in the skill, which is a common technique for instruction pinning but also a form of prompt injection.
Indirect Prompt Injection Surface: The skill instructs the agent to perform searches (glob, grep) within the node_modules/workflow/docs/ directory. This creates a surface where the agent could ingest instructions embedded in external documentation files.
Ingestion points: Markdown files located in node_modules/workflow/docs/ and associated package directories.
Boundary markers: Absent (There are no specific instructions or delimiters to tell the agent to treat the discovered documentation as non-executable data).
Capability inventory: The skill utilizes glob for file discovery, grep for content searching, and various npx workflow commands for system interaction.
Sanitization: Absent (The agent is directed to read and follow the content of the documentation files without validation or sanitization steps).
Command Execution via NPX: The debugging section lists several npx workflow commands. While these are intended for legitimate diagnostic purposes, using npx involves executing packages that may involve remote retrieval if not already present in the local cache.

workflow