The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute Python scripts where arguments such as {brand} and {channels} are interpolated directly from user-supplied input (e.g., python scripts/performance-monitor.py --brand "{brand}"). This pattern of dynamic command assembly can be vulnerable to command injection if input parameters are not strictly validated.
[PROMPT_INJECTION]: The skill performs indirect prompt injection as it ingests and processes untrusted data from external marketing channels.
Ingestion points: The skill fetches content from external sources via performance-monitor.py (ad creatives, email content), seo-executor.py (web page content for SEO audits), and ai-visibility-checker.py (mentions in AI search engines).
Boundary markers: The instructions do not define clear delimiters or specific instructions for the agent to ignore potentially malicious embedded commands within the fetched marketing data.
Capability inventory: The skill has access to the Bash tool, file read/write operations, and the ability to execute various data-pulling scripts.
Sanitization: There is no evidence of sanitization or filtering of the external content before it is incorporated into the audit reports or processed by the agent.
[SAFE]: The skill adheres to a read-only principle for all marketing platform interactions and restricts its internal data storage to application-specific directories (~/.claude-marketing/), avoiding access to sensitive system files or credentials.

campaign-audit