recoup-internal-marketing
Fail
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: HIGHDATA_EXFILTRATIONDYNAMIC_EXECUTIONREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill instructs the agent to read a sensitive credential file containing a Vercel authentication token from the local filesystem at '$HOME/Library/Application Support/com.vercel.cli/auth.json'. This token is subsequently used to authenticate API requests to Vercel's web analytics service.
- [DYNAMIC_EXECUTION]: The file 'references/publish-verify.md' contains instructions to execute a complex multi-line script via 'node -e'. This script performs dynamic module imports and interacts with the local filesystem to validate project configurations.
- [REMOTE_CODE_EXECUTION]: The skill performs remote dependency installation using 'npx', including fetching skills from an external BytePlus (ByteDance) repository and installing the ElevenLabs and Hyperframes packages.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from multiple social media platforms (X, YouTube, Instagram, LinkedIn) to guide marketing strategy and video generation.
- Ingestion points: Social media metrics and post data are fetched via the 'api.recoupable.dev' vendor endpoint in 'references/learn-from-socials.md'.
- Boundary markers: The instructions lack delimiters or explicit warnings to prevent the agent from following instructions embedded within the scraped social media content.
- Capability inventory: The skill has significant capabilities including executing shell commands, performing network operations, and writing to the local filesystem across its various scripts.
- Sanitization: The skill does not implement sanitization or validation of external content retrieved from social media profiles before incorporating it into the marketing workflow.
Recommendations
- AI detected serious security threats
Audit Metadata