team-mode
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes a local Python script
scripts/usage_by_model.pyto parse session logs and generate reports on token consumption and reasoning costs. - [INDIRECT_PROMPT_INJECTION]: The skill implements a subagent architecture where delegated agents ingest and process untrusted data from the workspace and external sources.
- Ingestion points: Subagents receive data through the
Sourcesfield in the dispatch brief and read local files during exploration or execution. - Boundary markers: The skill employs a structured dispatch contract and utilizes
fork_turns="none"to isolate child context. - Capability inventory: The system allows for workspace modifications through the
Executorrole and shell command execution for usage reporting. - Sanitization: The instructions mandate that the main thread must manually inspect all delegated artifacts, source changes, and diffs before accepting the work.
- [EXTERNAL_DOWNLOADS]: The setup documentation references the author's official GitHub repository (
github.com/oil-oil/codex-team-mode) as the source for obtaining canonical TOML configuration templates for the subagent profiles.
Audit Metadata