huawei-cloud-cdn-abnormal-status-code-analysis

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes CDN access logs, which contain untrusted data from external sources (such as User-Agent strings, Referer headers, and request paths). This creates a vulnerability surface for indirect prompt injection where malicious instructions embedded in log entries could attempt to influence the agent's behavior.\n
  • Ingestion points: The skill retrieves log files via hcloud CDN ShowLogs/v2 and processes them using scripts/fetch_cdn_log.py (referenced in SKILL.md and references/task-forensics.md).\n
  • Boundary markers: The agent is instructed to extract specific rows for analysis, but no explicit 'ignore instructions' delimiters are applied to the raw log content during interpolation into the prompt.\n
  • Capability inventory: The skill uses subprocess to run hcloud commands and executes a Python script with network access (requests library).\n
  • Sanitization: The fetch_cdn_log.py script performs field-level parsing and limits output to 200 lines by default, but it does not sanitize or escape the content of the log fields before presenting them to the AI agent.\n- [COMMAND_EXECUTION]: The skill's primary functionality relies on executing the hcloud CLI and a bundled Python script (scripts/fetch_cdn_log.py) to perform diagnostic queries and data processing. These operations are limited to read-only actions and are necessary for the skill's stated purpose.\n- [EXTERNAL_DOWNLOADS]: The Python log helper (scripts/fetch_cdn_log.py) downloads log files from URLs generated by the ShowLogs API. These URLs point to Huawei Cloud CDN's internal storage infrastructure.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 08:53 AM
Security Audit — agent-trust-hub — huawei-cloud-cdn-abnormal-status-code-analysis