The Agent Skills Directory

[PROMPT_INJECTION]: The instructions use authoritative language to command the agent to disregard its internal training data in favor of the provided text.
Evidence: "Your knowledge of workflow is outdated." and "Follow these instructions before starting on any workflow-related tasks:" in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow pattern that ingests untrusted user data into an AI agent context, creating a potential vulnerability to embedded malicious instructions.
Ingestion points: The myAgentWorkflow function in SKILL.md accepts a userMessage argument.
Boundary markers: Absent; the input is interpolated directly into the agent.stream messages array.
Capability inventory: The skill utilizes fetch for network requests and provides tools for data lookup and workflow orchestration.
Sanitization: Absent; there is no evidence of filtering or escaping the user-provided content.
[COMMAND_EXECUTION]: The skill documents the use of project-specific CLI tools for debugging and monitoring.
Evidence: Multiple npx workflow commands are listed for health checks, dashboard access, and run inspection.

workflow