The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to monitor external data sources including LinkedIn engagement, Slack/WhatsApp community messages, and third-party review platforms (G2, Capterra). This ingestion of untrusted content creates a surface for indirect prompt injection, where malicious instructions embedded in external reviews or messages could attempt to influence the agent's output during plan generation.
Ingestion points: Public LinkedIn engagement data (likes/comments), Slack community threads, and customer reviews from G2 and Capterra.
Boundary markers: The instructions do not specify the use of clear delimiters or warnings to treat external data as untrusted.
Capability inventory: The skill provides a script (scripts/check-output.py) for file validation and guides the agent in creating marketing deliverables. High-privilege capabilities such as network exfiltration or system modification are not present.
Sanitization: There is no explicit logic provided for sanitizing or filtering instructions that might be embedded in the processed external data.
[COMMAND_EXECUTION]: The skill includes a local Python validation script, scripts/check-output.py, and instructs the agent to execute it to verify the completeness and quality of the generated marketing plan. The script performs safe operations by reading a specified file and checking for required keywords and unresolved template placeholders.

customer-marketing