The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses a high-risk execution pattern where content fetched from remote, user-provided URLs is piped into a Python interpreter (python3). Automated security scans identified this as a remote code execution vector. While the instructions currently include a fixed script for data parsing using the -c flag, the structural pattern of piping untrusted network data to a command-line interpreter is a known attack vector.
[COMMAND_EXECUTION]: The skill logic requires executing shell commands (curl) to perform network operations and data retrieval. This provides the necessary capability for the intended documentation audit but increases the risk if the input URLs or parameters are maliciously influenced.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and processes large amounts of untrusted text from external websites.
Ingestion points: SKILL.md (Step 1 and Step 2) fetches documentation content from user-provided URLs via curl.
Boundary markers: Absent. The skill does not use XML tags, delimiters, or explicit instructions to the agent to disregard directives found within the external documentation.
Capability inventory: The agent utilizes bash_tool for network egress and local script execution, and has the ability to write HTML reports to the filesystem.
Sanitization: No validation or filtering is performed on the fetched content to remove or neutralize natural language instructions that might attempt to override the audit logic or manipulate the agent's behavior.

sdk-docs-auditor