NYC

debugging-dags

Warn

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • EXTERNAL_DOWNLOADS (MEDIUM): The skill uses uvx --from astro-airflow-mcp af to run commands. This mechanism downloads and executes a Python package from a remote registry (PyPI) at runtime. The package source 'astro-airflow-mcp' is not included in the pre-defined list of trusted organizations or repositories.\n- COMMAND_EXECUTION (LOW): The skill relies on shell command execution to perform all its tasks, including health checks, log retrieval, and clearing tasks. This is standard for its intended use case but remains a privileged action.\n- Indirect Prompt Injection (LOW): The skill is vulnerable to indirect prompt injection due to its processing of external logs.\n
  • Ingestion points: Data enters the context via af tasks logs and af dags errors in SKILL.md.\n
  • Boundary markers: Absent; there are no instructions to the agent to treat log content as untrusted or to use delimiters.\n
  • Capability inventory: The agent has the capability to execute arbitrary commands via the af CLI/uvx wrapper.\n
  • Sanitization: No evidence of sanitization or filtering of the log content before it is processed by the LLM.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 17, 2026, 04:47 PM