Verify Codex Exec
Purpose
Prove whether the current authenticated account can complete a real managed
codex_exec run. Loading this skill, reading configuration, or observing a
healthy service is not proof.
Required Check
- Confirm
codex_execexists in the current tool set. If it is absent, do not call another tool or simulate success. - Invoke
codex_execexactly once with:
{
"target": {
"kind": "managed_sandbox"
},
"workspace": {
"kind": "empty_git"
},
"prompt": "Reply with exactly CODEX_EXEC_READY",
"timeout_secs": 180
}
Do not alter the target, workspace, prompt, or timeout. Do not retry automatically. Do not ask the user for a token, agent key, credential, service ID, provider, model route, or sandbox setting.
Verdict
Return AVAILABLE only when all of these are true:
- the tool call completed;
- its result is a JSON object with
statusequal tosucceeded; targetequalsmanaged_sandbox;exit_codeequals0;- trimmed
outputequals exactlyCODEX_EXEC_READY.
Start the final answer with AVAILABLE, then explain in the user's language
that managed codex_exec works for this account. Include target and
elapsed_ms when present.
For every other result, never claim success:
- start with
UNAVAILABLEwhen the tool is absent, disabled, the user is ineligible, or credential readiness fails; - start with
INCONCLUSIVEfor timeout, malformed result, execution failure, or unexpected output.
Include the stable code or error_code and diagnostic_id when present.
Explain the category briefly in the user's language. Do not print a raw
upstream response when those stable fields are available.
Never expose access tokens, agent keys, request headers, credentials, or secret references.