claude-certification
Warn
Audited by Gen Agent Trust Hub on Aug 9, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill downloads Python scripts (
main.py) from the author's GitHub repository (rohitg00/ai-engineering-from-scratch) and executes them using the local Python interpreter. This allows for the execution of arbitrary code fetched from a remote source at runtime. - [COMMAND_EXECUTION]: The agent is instructed to run shell commands to execute Python code and run unit tests. Commands identified:
python3 <lesson-path>/code/main.pyandpython3 -m unittest discover -s <lesson-path>/code/tests -v. - [EXTERNAL_DOWNLOADS]: The skill fetches various resources from
raw.githubusercontent.com/rohitg00/ai-engineering-from-scratch/main/, including program policies, lesson content, quizzes, and executable code. - [DATA_EXFILTRATION]: The skill can perform network operations targeting the Anthropic Messages API. It accesses the
ANTHROPIC_API_KEYfrom the environment to authenticate these requests. While the skill includes instructions to avoid persisting or printing the key, the ability to read and use credentials for network requests is a sensitive capability. - [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted data and code from an external repository without explicit sanitization. Ingestion points: track JSON, lessons, and quizzes fetched from GitHub. Boundary markers: Absent. Capability inventory: Python execution, file writing (
CLAUDE-CERTIFICATION.md), and network API access. Sanitization: Absent.
Audit Metadata