Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →lingtai-ai/preset-skill-codex-pool
Official-source-led manual for the TUI `codex-pool` template, including the codex-auth-pool.json format and manual-edit protocol.
Passed 20 of 22 test cases with the skill, and 7 of 22 without it.
and 19 more in the full report
13 fixed7 passed either way2 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 13 test cases the skill fixed. Same model, same prompt, with and without the skill.
In `tui/internal/preset/preset.go:1476`, what specific backend endpoint URL and API key environment variable setting are configured for `codexPoolPreset`?
| Checked | Without skill | With skill |
|---|---|---|
| Uses https://chatgpt.com/backend-api/codex for the endpoint configuration. | Missed | Passed |
| Sets api_key_env to an empty string. | Missed | Passed |
The instructions your model reads when the skill loads.
Use this child for the named built-in codex-pool preset. codexPoolPreset in tui/internal/preset/preset.go:1476 mirrors codex: gpt-5.6-sol, https://chatgpt.com/backend-api/codex, empty api_key_env, thinking xhigh, web_search, and vision. Pooling changes OAuth account selection, not the model or endpoint. Keep provider codex-pool distinct from codex.
Use the official Codex models and authentication pages, then verify served acceptance against the same Codex route for each relevant pooled account. Do not use the standard OpenAI API catalog. The pool is not a model catalog. Do not turn one account's observation into a claim about every account.
Start at codexPoolPreset in tui/internal/preset/preset.go. Keep providerModels["codex-pool"] exactly aligned with providerModels["codex"], including gpt-5.6-sol, gpt-6-astra, gpt-5.6-terra, and gpt-5.6-luna in that default-first order, with matching modelHasVision entries and the Codex four-level codexThinkingOptions in tui/internal/tui/preset_editor.go. Recheck the constructor's vision, /codex endpoint, OAuth-only fields, and the pool store only when pool selection or file behavior is part of the change. Follow tui/internal/tui/SKILL.md and tui/CONTRACT.md.
The pool file is $LINGTAI_TUI_DIR/codex-auth-pool.json, or
~/.lingtai-tui/codex-auth-pool.json when that variable is unset. The pool
stores only refs and integer weights.
The flat shape is {"version": 1, "accounts": [{"path": "...", "weight": 1}]};
the classified shape is {"version": 2, "models": {"<exact model>": [...]}}.
Presence of the models key is what classifies the pool, not its size or
version. Classified pools refuse flat TUI edits with
errCodexPoolModelClassified. Invalid or missing pool data, or an unmatched exact model, falls back to the legacy single-account auth path. Weight 0 means the account is present but disabled. Selection happens at adapter/service construction, is weighted and sticky within one agent wake/session, and does not reselect an already-running session.
Configured weights are inputs, not measured shares, and the pool excludes molt_count.
test cases passed, vs 7 / 22 without
Published skills whose descriptions closely match this one, with their headline benchmark lift.
More skills from the same source and nearby in the registry, with their headline benchmark lift.