The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes audit event logs retrieved from the Volcengine platform, which may contain attacker-controlled strings (e.g., UserAgent, ResourceID, or Error Messages). This represents a theoretical surface for indirect prompt injection.
Ingestion points: Audit event data retrieved by scripts/cloud_trail.py via the Volcengine CLI's LookupEvents action.
Boundary markers: Instructions in SKILL.md and references/event-query.md direct the agent to transcribe logs using a structured JSONC format with field-level annotations.
Capability inventory: The skill possesses capabilities to create, modify, and delete audit trails and backfill delivery tasks.
Sanitization: The Python wrapper script parses the raw CLI output as structured JSON and selects specific keys for presentation, limiting the raw data passed directly to the agent.
[COMMAND_EXECUTION]: The skill uses a Python script (scripts/cloud_trail.py) to execute the official Volcengine command-line interface (ve).
Execution is performed using subprocess.run with argument lists rather than shell strings, mitigating shell injection vulnerabilities.
The script implements an allowlist of permitted cloud actions (ALLOWED_ACTIONS) and validates input parameters, such as time formats and pagination integers.
All write operations are protected by a pre-flight dry-run and require a user-confirmed --confirm flag before execution.
[EXTERNAL_DOWNLOADS]: The skill configuration specifies the installation of the @volcengine/cli Node.js package.
This package is the official tool provided by the vendor, Volcengine, and is downloaded from the standard npm registry.
Usage of this vendor-provided resource is consistent with the skill's primary purpose of cloud management.

volcengine-cloud-trail