The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes Huawei Cloud WAF security event logs, which inherently contain attacker-controlled input such as malicious payloads, custom headers, and User-Agent strings.
Ingestion points: Untrusted data is ingested via hcloud WAF ListEvent and hcloud WAF ShowEvent calls as defined in SKILL.md.
Boundary markers: The prompt does not specify boundary markers or explicit instructions for the agent to ignore instructions embedded within the logs.
Capability inventory: The agent is authorized to execute read-only hcloud commands and perform file operations (write/read/merge JSON) within the $env:TEMP directory.
Sanitization: No specific sanitization or filtering logic is defined for the log payloads before they are processed by the LLM.
Mitigation: The risk is effectively mitigated by the skill's core 'Read-Only Principle', which forbids the agent from executing any modifying commands (Create*, Update*, Delete*). The agent only generates copy-paste ready commands for the user, ensuring a human-in-the-loop for all security configuration changes.
[COMMAND_EXECUTION]: The skill uses the official hcloud CLI (Huawei Cloud Command Line Interface) to query security data. All commands executed by the agent are scoped to read-only operations (ListEvent, ShowEvent, ListPolicy, ShowPolicy, ListHost) against the vendor's own infrastructure. The use of --cli-jsonInput to handle user-supplied parameters like domains is a safe practice to prevent command injection.

huawei-cloud-waf-log-analysis