The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted research data to generate draft reports, which creates a surface for indirect prompt injection. Instructions hidden in research materials could potentially influence the drafting process. Ingestion points: the skill takes 'research findings' and 'collected research' as input via the /draft workflow. Boundary markers: the instructions do not define delimiters or specific 'ignore' directives for the ingested content. Capability inventory: the agent has the capability to write the generated output to the local 'papers/' directory. Sanitization: there is no mention of sanitizing or validating the input research data before it is processed by the writer and verifier agents.

paper-writing