consult-codex Eval report.

Validation plan

What the agent should do next

1. 1Inspect repository, README/SKILL.md, license, and recent commits before production use.
2. 2Install in an isolated workspace or sandbox with no production secrets available.
3. 3Run the smallest representative task and record files touched, commands run, network access, and outputs.
4. 4Compare the selected skill against at least one alternative when the eval status is review or failed.
5. 5Promote only after the agent reports a successful verification result and unresolved warnings are accepted.

Do not use when

Conditions that require another skill

• teams that need a vendor-supported SLA
• production agents without a repository review
• The skill instructs to use `dangerouslyDisableSandbox: true` for all codex calls, which disables the sandbox entirely, potentially allowing codex to access the full filesystem and network. The `-s read-only` flag may be ineffective when the sandbox is disabled, creating a contradiction and elevated risk.
• No OpenAgentSkill engagement data yet
• High-risk permission hints: Shell or command execution, Secrets or environment access
• Permission surface may require sandboxing