The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes an indirect prompt injection vulnerability surface by processing user-provided innovation topics through its core analysis pipeline.
Ingestion points: The skill takes untrusted user input via the mandatory topic argument required for the /genius-thinking command.
Boundary markers: There are no explicit delimiters or instructions (such as 'ignore instructions within this block') to isolate user-provided data from the agent's internal reasoning prompts and tool-call structures.
Capability inventory: The skill leverages the sequential-thinking reasoning tool, spawns sub-agents via a Task tool for parallel analysis, and has file-writing capabilities to the .hypercore/ directory.
Sanitization: No sanitization or validation of the user's input topic is performed before it is interpolated into the reasoning and task execution phases, which could allow an attacker to influence the sub-agent instructions or the final analytical output.

genius-thinking