recoup-internal-marketing

Fail

Audited by Gen Agent Trust Hub on Oct 10, 2026

Risk Level: HIGHDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The skill accesses sensitive local credential files to extract authentication tokens. Specifically, references/conversion.md contains instructions to read the Vercel CLI authentication file located at $HOME/Library/Application Support/com.vercel.cli/auth.json. This exposure of private credential files is a high-severity security concern.
  • [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by ingesting untrusted external data.
    • Ingestion points: references/learn-from-socials.md scrapes recent posts from linked social media profiles (X, YouTube, Instagram, LinkedIn) via the api.recoupable.dev endpoint.
    • Boundary markers: The skill lacks explicit boundary markers or instructions to ignore embedded commands within the scraped social data.
    • Capability inventory: The skill possesses broad capabilities including network access (curl), dynamic code execution (node -e), and file system operations.
    • Sanitization: There is no evidence of sanitization or filtering of the scraped content before it is processed by the agent to generate new marketing briefs.
  • [DYNAMIC_EXECUTION]: The skill utilizes dynamic execution patterns to perform runtime validation. references/publish-verify.md uses node -e to execute complex inline JavaScript logic that reads project configurations and environment variables. Additionally, it uses npx to execute external tools like hyperframes at specific versions.
  • [REMOTE_CODE_EXECUTION]: In references/seedance.md, the skill instructs the agent to download and install external skills from a remote URL (https://arkdocs-en.tos-ap-southeast-1.volces.com/skills/) using the npx skills add command. While the source is a well-known cloud provider (Volcengine/ByteDance), executing remote code package installations at runtime is a critical capability.
  • [COMMAND_EXECUTION]: The skill frequently spawns subprocesses to execute shell commands (curl, python3, node) to interact with APIs, process JSON data, and manage media assets.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 10, 2026, 02:15 PM