drama-quality-assessment

Pass

Audited by Gen Agent Trust Hub on Sep 8, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external analytics data from ThinkingEngine to perform quality assessments and generate AI-driven diagnoses. This represents an indirect prompt injection surface as the agent processes data from external systems.
  • Ingestion points: Data is retrieved via ae-cli commands in references/steps/02_query_data.md and processed by the AI in references/steps/04_diagnose.md.
  • Boundary markers: The skill implements strong process boundaries by requiring explicit user confirmation of the drama name and analysis scope in references/steps/01_parse_config.md before any data operations occur.
  • Capability inventory: The skill utilizes ae-cli for data retrieval and has the capability to export reports to Feishu documents using lark-doc and lark-base skills.
  • Sanitization: The diagnosis logic in references/steps/04_diagnose.md is strictly constrained by a local pattern library (references/anomaly_patterns.md), which limits the influence of external data on agent behavior.
  • [EXTERNAL_DOWNLOADS]: The skill references and interacts with ThinkingEngine (a well-known analytics service) and Feishu (Lark) for report generation. These integrations are essential to the skill's primary analytical purpose.
  • Evidence: The skill uses the ae-cli tool (adapter defined in references/adapters/ae-cli.md) to fetch drama performance metrics from ThinkingEngine projects.
  • Integrations: It provides optional export functions to Feishu (lark-doc, lark-base) as detailed in references/steps/05_generate_report.md.
  • [COMMAND_EXECUTION]: The skill uses ae-cli to execute data queries. These commands are constrained to analytical metadata and ad-hoc query models.
  • Evidence: Command patterns are strictly defined in references/adapters/ae-cli.md and involve operations like ae-cli analysis adhoc run and ae-cli analysis-meta event list.
  • Safety: The skill prohibits guessing command names or parameters, requiring they be sourced from technical metadata retrieved at runtime.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 8, 2026, 03:22 AM
Security Audit — agent-trust-hub — drama-quality-assessment