The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by design, as it processes untrusted user-supplied data to generate structured documents.
Ingestion points: The skill ingests study descriptions through user prompts and reads research specifications from local Markdown files via the --input argument in SKILL.md.
Boundary markers: There are no explicit boundary markers, delimiters, or system instructions defined to prevent the agent from obeying instructions that might be embedded within the input data.
Capability inventory: The skill utilizes the Read tool to access files, the Write tool to save drafts to the quality_reports/preregistrations/ directory, and the Agent tool to invoke sub-agents like /verify-claims and /power-analysis as specified in the Workflow sections of SKILL.md.
Sanitization: No sanitization, filtering, or escaping of the input data is performed before it is interpolated into the preregistration templates.

preregister