The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from client-provided materials (decks, brand docs, transcripts) and external APIs (Ahrefs, GA4, Stripe, Notion). This content is used to generate strategy documents and social media posts, creating a surface for indirect prompt injection attacks where malicious instructions hidden in the source data could be executed by the agent.\n
Ingestion points: ~/marketing-plans/{client-slug}/materials/ folder and data from Ahrefs, GA4, Stripe, and Notion MCPs.\n
Boundary markers: None identified; the skill does not explicitly use delimiters to separate ingested data from instructions.\n
Capability inventory: The skill can write files to the local filesystem, publish content to GitHub via CLI, and post to social media platforms via Typefully.\n
Sanitization: The skill performs a "verification pass" in Phase 3 to check for machine-specific paths, but it does not include explicit logic to sanitize or ignore instructions embedded in the ingested client data.\n- [COMMAND_EXECUTION]: The skill's documentation (references/ops-stack-mapping.md) recommends the manual installation of external CLI tools such as agent-browser and defuddle using npm. These tools are intended for use by the operator to perform browser-based automation tasks mentioned in the plan template.

marketing-plan