The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to run npx -y @nomadamas/k-skill@0, which downloads and executes a CLI tool from a remote registry. This tool is used to generate the final instructions and execute the skill's scripts, introducing an unverified remote execution layer.
[EXTERNAL_DOWNLOADS]: Operational components, including the primary instruction set and supplemental files, are retrieved from external repositories (GitHub and NPM) during the skill's execution cycle.
[COMMAND_EXECUTION]: The skill executes shell commands and Python scripts. User-provided keywords and URLs are passed as arguments to these scripts, increasing the risk of manipulation if inputs are not strictly validated.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from the Daangn marketplace. This is a vulnerability surface where a malicious seller could include hidden instructions in their product listings to influence the agent.
Ingestion points: Marketplace product titles and descriptions fetched via scripts/daangn_used_goods.py.
Boundary markers: The skill does not use specific delimiters or protective instructions to help the agent distinguish between data and commands.
Capability inventory: The agent has the ability to make network requests and execute shell scripts.
Sanitization: The script performs basic HTML unescaping but lacks validation to strip or sanitize potential prompt injection content.
[DATA_EXFILTRATION]: The detail command in the daangn_used_goods.py script accepts a URL parameter and fetches it directly. Because there is no check to ensure the URL belongs to daangn.com, an attacker could use this to perform SSRF attacks, potentially exposing internal metadata or fetching content from arbitrary external servers.

daangn-used-goods-search