The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs a network request to the developer's GitHub repository to retrieve the current version number. This is used solely to notify the user if a newer version of the tool is available.
[COMMAND_EXECUTION]: The orchestrator agent utilizes standard shell commands like find and cat to locate Solidity source files and prepare data bundles for analysis. These operations are restricted to the local file system and temporary directory.
[PROMPT_INJECTION]: The skill implements an audit workflow that involves ingesting untrusted Solidity source code into the agent's context for analysis. To mitigate the risk of the code being interpreted as instructions (indirect prompt injection), the skill employs boundary markers.
Ingestion points: Source files discovered via the find command and bundled in Turn 2 of the SKILL.md orchestration.
Boundary markers: Fenced code blocks and ### path headers are used to delimit external code from the agent's internal instructions.
Capability inventory: Local file system read access, network access for version checking, and the ability to spawn sub-agents for parallel analysis.
Sanitization: The skill wraps untrusted code in markdown structures to maintain clear context separation, following common safety patterns for code-analysis agents.

solidity-auditor