The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest external content, such as web page markdown and public reviews, using the sandbase_run tool to perform live research. This creates an attack surface where malicious instructions embedded in third-party websites or reviews could potentially influence the agent's behavior during the drafting process. \n- Ingestion points: External data is fetched via the sandbase_run tool (e.g., scrape markdown, google organic serp, public review source) as described in SKILL.md and references/sandbase-api-map.md. \n- Boundary markers: The skill does not define specific delimiters for separating external data from instructions, though it advises the agent to quote only the minimum necessary and prefer paraphrase. \n- Capability inventory: The agent has the capability to read external web data, write the final article, and save screenshots to local folders. \n- Sanitization: The instructions do not mention any sanitization, filtering, or validation of the ingested external content before it is processed by the model.

listicle-blog-writer