codex
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill makes extensive use of the
Bashtool to executegitcommands for repository state detection and diff extraction. It also runs several local utility binaries from thegstackframework (e.g.,gstack-skill-start,gstack-codex-probe,gstack-review-log) for telemetry, session management, and environment verification. - [EXTERNAL_DOWNLOADS]: The skill recommends the installation of the
@openai/codexCLI tool from the official npm registry, which is a well-known service. This is required for the skill's primary functionality of performing external code reviews. - [INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted data such as
git diffoutput and project plan files. To mitigate injection risks, it usesDIFF_STARTandDIFF_ENDdelimiters to wrap diff data and prepends a 'Filesystem Boundary' instruction to all prompts sent to the external model, explicitly forbidding access to agent configuration and skill files. - [DYNAMIC_EXECUTION]: The skill executes hardcoded, inline Python code to parse JSONL output from the Codex CLI. It also uses
bunto run local TypeScript logic (outside-review-result.ts) for processing review outcomes. These executions are scoped to data processing and do not include user-supplied code snippets.
Audit Metadata