The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's workflow involves reading and processing external research papers, ideas, and code repositories. This provides a vector for indirect prompt injection, where malicious instructions hidden within research documents could potentially influence the agent's output during the planning phase.
Ingestion points: The skill ingests user input via the $0 argument and external files during the 'Understand the Research Context' step described in SKILL.md and referenced in planning-prompts.md.
Boundary markers: While the prompt templates in planning-prompts.md use structural headers like '## Research Context', they do not include explicit instructions to ignore or disregard potentially malicious commands embedded in the source research material.
Capability inventory: The skill focus is on generating structured text artifacts (JSON, Mermaid, markdown) and does not define access to dangerous tools like remote command execution or file-system writing in its configuration.
Sanitization: The skill does not perform any validation or sanitization of the research content before processing it to generate methodology plans.

research-planning