The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes untrusted Solidity source code, which could contain comments or strings with prompt injection payloads intended to influence the AI agents.
Ingestion points: The skill reads project Solidity files using shell commands like find and cat.
Boundary markers: Ingested files are wrapped in headers and markdown code blocks in the temporary audit bundle.
Capability inventory: The orchestrator executes shell commands and calls multiple external AI agents with different specialty prompts.
Sanitization: The skill does not filter input code for injection payloads, relying instead on agent instructions to operate in a read-only capacity.

solidity-auditor