The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the agent to clone and install a toolkit from an untrusted GitHub repository: https://github.com/fairy-root/drm-pentesting-toolkit.git.
[DYNAMIC_EXECUTION]: The toolkit's architecture relies on dynamically loading and executing Python scripts stored in the plugins/ directory, which could allow for the execution of arbitrary code if malicious files are placed there.
[COMMAND_EXECUTION]: The skill uses subprocess.Popen to run external executables (e.g., N_m3u8DL-RE). The arguments for these commands are built from potentially untrusted external data like manifest URLs and extracted keys.
[CREDENTIALS_UNSAFE]: The toolkit is designed to manage and expose highly sensitive data, including Widevine/PlayReady device files containing private keys, and user authentication credentials (passwords, API keys) stored in local text files.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests data from external DASH/MPD manifests and license server responses via requests calls.
Boundary markers: No specific boundary markers or sanitization logic is described for handling these external data sources.
Capability inventory: The skill has capabilities for network requests (requests), local file writing, and shell command execution (subprocess).
Sanitization: The extraction logic for PSSH and keys uses simple regular expressions and base64 decoding, which might be vulnerable to malicious payloads embedded in manifests.
[PRIVILEGE_ESCALATION]: The toolkit includes instructions and code snippets for programmatically changing file permissions (chmod +x) to allow the execution of external binaries.

drm-pentesting-toolkit