The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The documentation instructs users to install the SquirrelScan CLI by piping remote scripts to the shell via curl -fsSL https://squirrelscan.com/install | bash (macOS/Linux) and irm https://squirrelscan.com/install.ps1 | iex (Windows). These patterns execute opaque remote code without local verification.
[EXTERNAL_DOWNLOADS]: The skill requires downloading external binaries and configuration scripts from squirrelscan.com. While associated with the tool, these are not from pre-verified trusted organizations.
[COMMAND_EXECUTION]: The skill relies on the squirrel binary and suggests the agent execute local development servers and formatting tools like ruff, biome, and tsc to apply fixes found during audits. This grants the agent significant execution capabilities based on audit results.
[DATA_EXFILTRATION]: Instructions direct the agent to search environment variables (e.g., linked Vercel projects) and the local directory to find audit targets, which could inadvertently expose internal configuration data.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes content from arbitrary external websites. Ingestion points: Remote website HTML, metadata, and link structures. Boundary markers: The skill uses a custom XML/text report format but lacks explicit instructions to the agent to disregard commands embedded in the audited content. Capability inventory: Access to the Bash tool and instructions to modify local code to implement audit fixes. Sanitization: No sanitization of the scraped content is performed before presentation to the agent.

audit-website