The Agent Skills Directory

[OBFUSCATION]: Detection of a typosquatted URL within the skill documentation. The skill is authored by 'mabzadev' and identifies its components using the 'mabza' prefix. However, a link in SKILL.md directs the agent to https://github.com/mbzadev/skills/blob/main/tools/REGISTRY.md to retrieve tool integration guides. The omission of the letter 'a' in the GitHub organization name is a classic indicator of impersonation and supply chain attacks intended to deceive users or automated tools into accessing unauthorized repositories.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a natural language router that processes untrusted descriptions of user situations to trigger other tool workflows.
Ingestion points: User situation descriptions processed in SKILL.md to determine the appropriate skill or workflow.
Boundary markers: No specific delimiters or safety instructions (e.g., "ignore embedded instructions") are present to wrap the user-supplied data.
Capability inventory: The system supports high-privilege capabilities including command execution (/implement, /tdd), background network research (/research), and writing to local environment files (.env) and diagnostic logs.
Sanitization: There is no evidence of input validation or sanitization, allowing potential malicious instructions in the user's situation description to influence the execution of downstream skills.

ask-mabza