The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation instructs the agent or user to clone an external repository and execute a shell script (bash scripts/install.sh). Running unverified scripts downloaded from third-party sources is a high-risk operation that could lead to unauthorized code execution.
[COMMAND_EXECUTION]: The skill requires various shell commands for setup and management, including file operations (mkdir, cp, chmod) and the execution of conversion tools (scripts/convert.sh).
[EXTERNAL_DOWNLOADS]: To function, the skill must download framework definitions and scripts from a GitHub repository (https://github.com/BrianRWagner/ai-marketing-claude-code-skills.git) managed by an external author.
[PERSISTENCE]: The skill guides users to store API keys for external services in shell profile files like ~/.bashrc or ~/.zshrc. This ensures the credentials persist across sessions but exposes them to any local process capable of reading these files.
[INDIRECT_PROMPT_INJECTION]: Features such as last30days and reddit-insights ingest content from platforms like Reddit, X/Twitter, and YouTube. This introduces a vulnerability where malicious instructions embedded in the external content could manipulate the agent's behavior.
Ingestion points: Processes content from Reddit, X/Twitter, and YouTube transcripts via web search and API calls.
Boundary markers: The framework description does not mention the use of delimiters or specific instructions to ignore adversarial prompts in the fetched data.
Capability inventory: The skill has permissions to execute shell scripts, write to the file system, and initiate network requests.
Sanitization: The documentation provides no information regarding the validation, filtering, or escaping of data retrieved from external sources before it is analyzed by the AI model.

ai-marketing-claude-code-skills