coupang-product-search
Fail
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill instructs the agent to run
npx -y @nomadamas/k-skill@0, which downloads and executes code from an external, untrusted package registry. This execution is used to fetch the skill's primary instructions and associated files at runtime, allowing the skill's behavior to be changed dynamically without a static update to the skill itself. - [COMMAND_EXECUTION]: The skill requires the execution of multiple shell commands (
npx,curl) to retrieve instructions, legal disclaimers, and search data from remote services. These commands are integral to the skill's operation but introduce risks related to the integrity of the downloaded content. - [INDIRECT_PROMPT_INJECTION]: The skill processes product data retrieved from an external proxy (
k-skill-proxy.nomadamas.org) and uses this information to guide the agent through browser-based commerce workflows, including purchasing and cart management. This creates a vulnerability where malicious upstream data could manipulate agent behavior. - Ingestion points: Product search results are ingested from the proxy endpoint
v1/coupang/products/search(referenced ininstruction.md). - Boundary markers: No specific delimiters or boundary markers are defined to isolate untrusted product metadata from the agent's internal reasoning or decision-making processes.
- Capability inventory: The skill possesses extensive capabilities including web browsing, automated commerce actions (cart/purchase), and access to a credential vault via
request_vault_credential. - Sanitization: There is no evidence of sanitization or validation logic to ensure that product titles, descriptions, or URLs do not contain malicious instructions intended to hijack the agent's workflow.
- [METADATA_POISONING]: The skill metadata and instructions describe the proxy server as an "official" path for the Coupang Partners API, which contradicts the explicit legal disclaimers stating that the skill has no official affiliation or sponsorship from the service provider. This discrepancy could mislead users or agents regarding the source and security of the proxy service.
Recommendations
- AI detected serious security threats
Audit Metadata