The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to clone a repository from a personal GitHub account (https://github.com/ericosiu/ai-marketing-skills.git) rather than a recognized trusted organization.
[REMOTE_CODE_EXECUTION]: The installation process involves running pip install -r requirements.txt against the downloaded repository, which may result in the installation of unverified dependencies or the execution of arbitrary code.
[COMMAND_EXECUTION]: The skill documentation provides numerous examples of CLI commands (e.g., python experiment-engine.py, python pacing-alert.py) that the agent is meant to execute. These scripts interact with external APIs and local file systems.
[INDIRECT_PROMPT_INJECTION]: The skill creates a broad attack surface by ingesting data from external sources such as website visitor intent, sales call transcripts, and content files for processing by an LLM.
Ingestion points: The skill fetches visitor data via the RB2B API, sales call insights from the Gong API, and reads local markdown files (blog-post.md) for quality scoring.
Boundary markers: There are no documented boundary markers or specific instructions for the LLM to ignore potentially malicious embedded commands within the ingested data.
Capability inventory: The associated scripts have capabilities to perform network requests, write data to CRMs (Salesforce), and generate outbound communications.
Sanitization: The documentation references a security/sanitizer.py utility and sanitize_output function designed to remove PII and API keys, though the effectiveness depends on the implementation within the external repository.

ai-marketing-skills-automation