The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is vulnerable to a Local File Read attack. The curl --data-urlencode command in the Python script directly interpolates the user's address query. By providing an input starting with the @ symbol (e.g., @/etc/passwd), an attacker can force curl to read local files and send their contents to the official ePost server.\n- [COMMAND_EXECUTION]: The skill uses subprocess.run to execute the curl command-line tool from within a Python environment.\n- [REMOTE_CODE_EXECUTION]: The skill employs a dynamic execution pattern by wrapping a Python script in a shell here-doc. While the code is provided within the skill, this method of execution represents a higher risk profile for code injection if combined with other vulnerabilities.\n- [DATA_EXFILTRATION]: The skill exhibits an Indirect Prompt Injection surface. It fetches and processes HTML data from an external official source and presents it to the agent for normalization without sufficient verification.\n
Ingestion points: HTML response from parcel.epost.go.kr in SKILL.md.\n
Boundary markers: Absent.\n
Capability inventory: subprocess.run and curl network access.\n
Sanitization: Limited to html.unescape for display purposes.

zipcode-search