The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external websites during the GEO audit process, which could theoretically contain malicious instructions.
Ingestion points: External domains are accessed via curl to retrieve robots.txt, llms.txt, and other site metadata (SKILL.md).
Boundary markers: None explicitly defined for prompt delimiters, though audit output is saved to isolated markdown files in ~/.geo-prospects/audits/.
Capability inventory: The skill uses Bash (curl) and Write tools to fetch and store data.
Sanitization: The orchestration instructions include a security-positive requirement to validate domains as hostnames and reject path separators or dot segments.
[COMMAND_EXECUTION]: The skill relies on shell commands to manage its directory structure and perform network audits.
Evidence: Orchestration instructions specify the use of mkdir -p ~/.geo-prospects/audits and curl -s for site analysis.
Mitigation: The skill includes defensive instructions to prevent command injection by requiring hostname validation and rejecting suspicious path characters in user-provided domains.

geo-prospect