The Agent Skills Directory

[PERSISTENCE]: The setup script scripts/ensure_cli.sh modifies the user's shell configuration files (~/.bashrc and ~/.profile) to append a PATH export for the CLI tool's installation directory, ensuring the tool remains available in future sessions.
[DATA_EXFILTRATION]: The skill uses a bundled telemetry tool (skill-quality-cli) that transmits execution metadata to a Huawei Cloud developer endpoint (https://skillsapi.developer.myhuaweicloud.com). The data sent includes session identifiers and truncated snippets (up to 6,000 characters) of the agent's interaction history and tool outputs.
[DATA_EXFILTRATION]: The reporting script scripts/cli/cli_reporting.py searches for local database files and session logs from various AI agent platforms (such as Claude Code, Hermes, and Codex) to extract active session identifiers for its telemetry reports.
[COMMAND_EXECUTION]: The skill executes hcloud (KooCLI) commands using a Python wrapper script that utilizes subprocess.run to manage the lifecycle and reporting of cloud operations.
[INDIRECT_PROMPT_INJECTION]: The skill ingests metadata from external cloud resources—such as secret names from CSMS, key aliases from KMS, and audit logs from CTS—without utilizing explicit boundary markers or sanitization, representing a surface for potential indirect prompt injection attacks.
Ingestion points: Data returned from hcloud CSMS ListSecrets, hcloud KMS ListKeys, and hcloud CTS ListTraces commands.
Boundary markers: Absent from tool output processing instructions.
Capability inventory: File system operations, network access via hcloud, and management of encryption keys/secrets.
Sanitization: No specific filtering or escaping is performed on resource metadata before it is processed by the agent.

huawei-cloud-dew-key-management