turbo
Warn
Audited by Gen Agent Trust Hub on Jul 16, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes the
mcp__speed-run__generate_and_write_filestool to write content generated by an external LLM provider directly to the local filesystem. This process is designed to bypass the standard interaction where the agent displays code for user approval before writing to disk. - [REMOTE_CODE_EXECUTION]: The proposed workflow involves generating source code through the Cerebras API and immediately running a test suite against it. This sequence effectively results in the execution of code generated by a remote, untrusted source.
- [PROMPT_INJECTION]: The skill processes a 'contract prompt' as its primary input. This serves as an ingestion point for untrusted data that influences the behavior of the remote LLM, creating a surface for indirect prompt injection that could result in the generation of malicious payloads.
- [CREDENTIALS_UNSAFE]: The skill relies on the
CEREBRAS_API_KEYenvironment variable. While required for the service's functionality, it introduces a requirement for managing sensitive API credentials within the agent's execution environment.
Audit Metadata