The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a Python script (local-seo-checker.py) with arguments constructed from user-provided Name, Address, and Phone (NAP) data. This pattern is vulnerable to command injection if inputs are not strictly sanitized before being passed to the shell execution environment.
Evidence: python "${CLAUDE_PLUGIN_ROOT}/scripts/local-seo-checker.py" --nap '{"name":"...","address":"...","phone":"..."}' --industry {industry}
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from user inputs (Business name, Website URL) and external sources (Google Business Profile, Citation sites). This content is interpolated into the agent's context without explicit boundary markers or sanitization, creating a surface for indirect instructions to influence agent behavior.
Ingestion points: SKILL.md (Business name/address, Website URL, GBP URL)
Boundary markers: Absent
Capability inventory: Python script execution, file reading from ~/.claude-marketing/, report generation.
Sanitization: Not explicitly mentioned in the process.
[DYNAMIC_EXECUTION]: The skill utilizes a local Python script to perform core logic, including the dynamic processing of brand-specific JSON profiles and marketing guidelines stored in the user's home directory.
Evidence: Accesses ~/.claude-marketing/brands/ to load profile and manifest data.

local-seo-audit