ks-ai-feed

Fail

Audited by Gen Agent Trust Hub on Aug 24, 2026

Risk Level: HIGHCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PERSISTENCE_MECHANISMS]: The install_subscription function in assets/daily_report.py creates a persistent LaunchAgent on macOS (com.qoder.ks-ai-feed.plist) and modifies the crontab on Linux and Windows systems. This allows the skill's scripts to execute automatically in the background at scheduled times, maintaining presence across sessions.\n- [DATA_EXPOSURE_AND_EXFILTRATION]: The ProxyHTTPHandler class in assets/daily_report.py implements an image proxy endpoint (/api/img) that fetches content from arbitrary URLs provided in query parameters using urlopen. Because the destination domain is not validated, this constitutes a Server-Side Request Forgery (SSRF) vulnerability that could be used to probe the local network or exfiltrate data through the local server.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted video metadata (titles, tags, and author names) from the Kuaishou platform and instructs the agent to process and summarize this information. \n
  • Ingestion points: Video data fetched by fetch_articles in assets/daily_report.py.\n
  • Boundary markers: Absent; external content is interpolated directly into report templates and terminal outputs without delimiters.\n
  • Capability inventory: Subprocess execution for system configuration, local HTTP server operations, and file writing to the user's Downloads directory.\n
  • Sanitization: None; raw strings from external sources are processed without escaping or filtering.\n- [COMMAND_EXECUTION]: The script uses subprocess.run to execute shell commands for installing persistence mechanisms and opening system browsers. The use of shell=True when modifying the crontab involves string interpolation of file paths, which presents a security risk.\n- [EXTERNAL_DOWNLOADS]: The skill fetches reporting data and intelligence configuration from the vendor domain redfox.hk. These operations are consistent with the skill's primary functionality.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Aug 24, 2026, 06:06 PM
Security Audit — agent-trust-hub — ks-ai-feed