The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (HIGH): The skill is designed to fetch resources from external URLs using Invoke-WebRequest and curl. There is no whitelist of domains or validation of the file types being downloaded, allowing an attacker to trick the agent into downloading malicious scripts or binaries.
[COMMAND_EXECUTION] (HIGH): Instructions in SKILL.md and scripts.md provide direct command-line templates for network and filesystem operations. These commands are executed with the agent's privileges, which could be abused to access sensitive files if the agent is misled by an indirect prompt injection.
[INDIRECT_PROMPT_INJECTION] (HIGH): The skill processes untrusted data (URLs and external files) and possesses high-impact capabilities (network access and filesystem writes).
Ingestion points: User-provided URLs and resource metadata from the web are ingested into the agent's context in SKILL.md and scripts.md.
Boundary markers: Absent. There are no delimiters or instructions to ignore embedded commands within downloaded content or metadata.
Capability inventory: System command execution (curl, PowerShell), filesystem modification (New-Item, Out-File, Rename-Item), and network requests are present across all script files.
Sanitization: Absent. No validation or filtering is performed on the URLs, filenames, or file content before processing.
[REMOTE_CODE_EXECUTION] (MEDIUM): Although the primary focus is on media resources, the downloader functions in scripts.md are generic. An attacker could provide a URL to a malicious script, which the agent would then download to the local machine, posing a significant risk of subsequent execution.

resource-downloader