twitter-downloader
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/download_tweet.pyexecutes external commands to check for and run the downloader. - Evidence: The script uses
subprocess.runandsubprocess.Popento callyt-dlpwith arguments like--versionand download parameters. - Security: Commands are passed as lists (e.g.,
['yt-dlp', '-g', url]) rather than strings, which correctly avoids the use of a shell and prevents shell injection vulnerabilities from user-supplied URLs. - [EXTERNAL_DOWNLOADS]: The skill requires the installation of an external third-party package.
- Evidence:
SKILL.mdinstructs users to installyt-dlpvia pip and links to its official GitHub repository. - Context:
yt-dlpis a highly reputable, widely-used open-source media downloader. - [INDIRECT_PROMPT_INJECTION]: The skill processes data from external URLs, creating a surface for potential indirect instructions embedded in third-party content.
- Ingestion points: The
urlargument inscripts/download_tweet.pyaccepts external X/Twitter post links. - Boundary markers: The skill does not use specific LLM boundary markers for the URL content, but the data is passed to a specialized CLI tool rather than back into an LLM prompt.
- Capability inventory: The skill has file-system write access and the ability to execute the
yt-dlpcommand. - Sanitization: The script implements regex validation (
(x\.com|twitter\.com)/\w+/status/\d+) to ensure input URLs conform to expected Twitter post formats before processing. - [METADATA_POISONING]: There is a minor inconsistency in the skill's documentation regarding its dependencies.
- Evidence: The title and description in
SKILL.mdrefer to a tool namedtwmd, while the usage instructions, dependencies, and implementation script useyt-dlp. - Assessment: This appears to be a benign documentation error or an artifact of a tool change during development, posing no security risk to the user.
Audit Metadata