The Agent Skills Directory

[COMMAND_EXECUTION]: The 'Memory Protocol' section in SKILL.md instructs the agent to execute shell commands (e.g., cat .claude/context/memory/learnings.md) to retrieve state before starting and to update these files after completion. This grants the agent specific file-system access patterns beyond simple tool use.
[PROMPT_INJECTION]: The skill's core functionality involves processing untrusted content through templates in SKILL.md (e.g., First Principles, Socratic Questioning) using the {content} variable. This presents a surface for indirect prompt injection where malicious instructions within the content could subvert the reasoning process. 1. Ingestion points: {content} placeholder in multiple templates within SKILL.md. 2. Boundary markers: Triple-dash delimiters (---) are used to wrap content. 3. Capability inventory: The skill has access to Read and Write tools and the ability to execute shell commands via the memory protocol. 4. Sanitization: There is no evidence of sanitization or specific instructions for the agent to ignore directives found within the processed content.
[PROMPT_INJECTION]: The file commands/advanced-elicitation.md contains a forceful directive ('follow it exactly as presented to you') which is a pattern often associated with attempting to override previous instructions or system safety protocols.

advanced-elicitation