run
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill manages a multi-agent workflow that ingests data from the user and the repository environment, which is then fed into several sub-agents. This structure creates an attack surface where malicious content within the repository or the task description could influence the behavior of the delegated agents.
- Ingestion points: The user's request (task), the
.codex/dev-team.mdconfiguration file, and existing repository source code in theREPOSITORYdirectory. - Boundary markers: The skill utilizes fresh contexts for agents (
fork_context: false) and isolates stage outputs into separate markdown files (e.g.,brief.md,plan.md). - Capability inventory: The skill executes local Node.js scripts (e.g.,
run-state.mjs) via subprocess calls, performs Git commands (git status), and writes files to the local run directory. - Sanitization: There is no explicit description of sanitizing external input or repository content before it is processed by the agents.
- [COMMAND_EXECUTION]: The skill executes local scripts as part of its core development orchestration logic.
- Runs
node "PACKAGE/scripts/doctor.mjs"to verify the environment. - Runs
node "PACKAGE/scripts/run-state.mjs"to initialize the run, record stage outcomes, and manage the execution state. - Runs
node "PACKAGE/scripts/codex-audit.mjs"when process-based auditing is selected by the user.
Audit Metadata