The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes CDN access logs retrieved from remote URLs. These logs contain untrusted end-user data such as request paths, User-Agent strings, and Referer headers which could be manipulated by an attacker to influence the agent.
Ingestion points: The scripts/fetch_cdn_log.py script downloads and decompresses CDN log files from presigned URLs generated by the hcloud CDN ShowLogs/v2 command.
Boundary markers: The script parses logs into structured JSON fields (e.g., client_ip, status, user_agent) but also includes the raw log line in its output. The instructions do not contain specific delimiters or warnings to ignore instructions embedded within this data.
Capability inventory: The agent has access to the hcloud (KooCLI) command-line interface. Although the instructions provide a comprehensive list of prohibited operations to enforce a read-only workflow, the underlying tool access is broad.
Sanitization: Log parsing is performed using standard Python modules (shlex, re) and output via json.dump. No semantic sanitization or filtering of potentially malicious instructions within the log fields is performed.
[COMMAND_EXECUTION]: The skill relies on the execution of multiple hcloud CLI commands to query Huawei Cloud CDN statistics, configurations, and logs. Additionally, it executes a local Python utility, scripts/fetch_cdn_log.py, to process the downloaded log files.
[EXTERNAL_DOWNLOADS]: The skill requires the requests Python library (version 2.25 or higher) and provides instructions to install it using pip. This is a standard, well-known package from the official Python package registry.

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