gpt-orchestration-build
Pass
Audited by Gen Agent Trust Hub on Jul 10, 2026
Risk Level: SAFENO_CODEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [SAFE]: No malicious patterns or security vulnerabilities were identified in the instructions or configuration files.
- [NO_CODE]: This skill contains only instructional text and metadata; no scripts, executables, or automated logic are included.
- [PROMPT_INJECTION]: The skill processes untrusted external data (findings, reports, issues) which is a potential surface for indirect prompt injection. Mitigation: It includes robust instructions for diff inspection and scope limitations. 1. Ingestion points: Finding ledger gathered from conversation, reports, plans, and issues. 2. Boundary markers: Data is organized into a formal ledger structure. 3. Capability inventory: Writing to files and executing build/test commands. 4. Sanitization: Mandatory inspection of generated diffs and scripts before execution.
- [COMMAND_EXECUTION]: The instructions involve executing verification commands and build scripts. Mitigation: It explicitly warns the agent to inspect scripts before running and to avoid production environments by default.
Audit Metadata