The Agent Skills Directory

[PROMPT_INJECTION]: The skill operates by ingesting and processing untrusted external data, which creates a surface for indirect prompt injection attacks. According to the evidence chain:
Ingestion points: The skill reads files from a user-provided materials/ directory and fetches data from external APIs including Ahrefs, Google Analytics, Stripe, Customer.io, and Shopify.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions to prevent the agent from following malicious commands embedded in the ingested data.
Capability inventory: The skill has significant capabilities, including reading and writing multiple local files (sections/*.md, research.md, final_plan.md) and performing Git operations (clone, commit, push) to publish data to external GitHub repositories.
Sanitization: No explicit sanitization, validation, or escaping of external content is documented before it is interpolated into the agent's context or included in the final output.
[EXTERNAL_DOWNLOADS]: The skill's operational documentation references the installation and execution of external tools such as agent-browser and defuddle via npm install, which are used to support its research and automation functions.
[COMMAND_EXECUTION]: The skill includes functionality to publish finalized documents to a remote GitHub repository, requiring the execution of shell commands for git clone, git commit, and git push.

marketing-plan