The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute a local script scripts/guidelines-manager.py with arguments {name} and {content} provided by the user. If the underlying platform executes this via a shell without strict escaping, it may be vulnerable to command injection through maliciously crafted SOP content.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text from users (process descriptions, workflow steps) and saves it to the local filesystem (~/.claude-marketing/sops/). This content is intended to be referenced by other modules to guide future agent actions, creating a surface for indirect prompt injection if the SOPs contain instructions designed to override agent behavior.
Ingestion points: User-provided SOP content and names processed in SKILL.md.
Boundary markers: No specific delimiters or safety warnings are instructed for the saved SOP content to prevent it from being interpreted as instructions by the LLM later.
Capability inventory: File system writing and execution of the guidelines-manager.py script.
Sanitization: The process lacks explicit sanitization or validation steps for the user-supplied content before it is structured and saved.
[DYNAMIC_EXECUTION]: The skill uses dynamic context and string interpolation to build command-line calls (e.g., guidelines-manager.py --action save-sop --name {name} --content "{content}"). The use of unvalidated, user-controlled variables in execution contexts is a significant security risk.

import-sop