quark-search
Fail
Audited by Gen Agent Trust Hub on Feb 20, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- COMMAND_EXECUTION (HIGH): The skill constructs shell commands by interpolating variables like
kw(search keyword) andpwd_id(resource ID) directly intocurlcommand strings (Steps 1, 2, and 5). Since these variables are sourced from user input or untrusted third-party API results (PanSou) and are not shell-escaped or quoted, a malicious input or API response could include shell metacharacters (e.g.,&,;,$()) to execute arbitrary commands on the host machine.\n- REMOTE_CODE_EXECUTION (HIGH): The skill pipes output from a local application service (localhost:9128) directly topython3. While the command specifically uses-m json.toolfor formatting, an automated scanner flagged this as a remote code execution pattern. If an attacker can influence the local service's output, they could potentially exploit the shell interaction or the interpreter.\n- EXTERNAL_DOWNLOADS (MEDIUM): The skill interacts withs.panhunt.com, an unverified third-party search aggregator, anddrive-pc.quark.cn. It relies on these external sources to provide data that controls subsequent local execution parameters and user-facing links.\n- DATA_EXFILTRATION (LOW): The skill accesses local application status (isLogin) and cloud drive metadata. While used for legitimate functionality, this information exposure is a prerequisite for more advanced attacks, and the local service at port 9128 is accessed without authentication.\n- PROMPT_INJECTION (LOW): The skill is vulnerable to Indirect Prompt Injection (Category 8).\n - Ingestion points: Search results from the PanSou API (
s.panhunt.com) in Step 1.\n - Boundary markers: Absent; the agent is instructed to directly parse and act upon data from external JSON responses.\n
- Capability inventory: Execution of
curlandopen(browser) commands via shell, and interaction with a local unauthenticated service.\n - Sanitization: Absent; the skill suggests using
strict=Falsewhen parsing JSON from the external API, which reduces the agent's ability to detect malformed or malicious payload structures.
Recommendations
- HIGH: Downloads and executes remote code from: http://localhost:9128/desktop_info - DO NOT USE without thorough review
- AI detected serious security threats
Audit Metadata