Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →lingtai-ai/preset-skill-op-endpoint-capabilities
Base URL / API-compat / provider / model / capability declaration shape versus credentials and live probes; includes proven Codex OAuth quota inspection with exact agent query routing and dated official-vs-measured context-window evidence.
Passed 21 of 21 test cases with the skill, and 10 of 21 without it.
and 18 more in the full report
12 fixed10 passed either way0 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
A newer attempt on gemini-3.6-flash on 10/2/2026 did not produce a publishable comparison (not comparable). The snapshot above is the previously measured run.
One of 11 test cases the skill fixed. Same model, same prompt, with and without the skill.
Before sending an `account/rateLimits/read` request to the Codex app-server, what initial handshake step must the client complete?
| Checked | Without skill | With skill |
|---|---|---|
| States that the client must complete the initialize handshake with clientInfo first. | Missed | Passed |
The instructions your model reads when the skill loads.
Every built-in preset's manifest.llm declares a fixed set of fields; this
child names what each one is and is not, so provider children can link here
instead of re-explaining the shape every time.
provider — one of the four families the kernel accepts: openai,
anthropic, codex, claude-code. It selects the adapter; it is not
itself a credential. Vendor-named providers (custom, deepseek,
minimax, openrouter, ...) are rejected by the kernel; other vendors are
reached through openai/anthropic with their compatible base_url.model — the exact model string sent to the endpoint. codex and
claude-code use curated picker catalogs (Codex OAuth models, Claude CLI
aliases); openai and anthropic are free text because the endpoint is
arbitrary.base_url — the endpoint the adapter calls. Optional for openai
(official https://api.openai.com/v1, SDK-style with its own version path)
and anthropic (official https://api.anthropic.com, /v1/... appended);
codex uses its /backend-api/codex route; claude-code has none.wire_api — openai family only: chat_completions (default,
written explicitly) or responses; responses_transport (http default,
or websocket) applies only to the Responses wire. The retired
api_compat field is no longer read by the kernel and is never written by
the TUI.thinking / service_tier — standard reasoning effort for openai
and anthropic (none..xhigh, omitted = kernel default); Codex keeps its
low..xhigh ladder with an explicit xhigh default. service_tier
(fast → priority) applies to openai and codex only.capabilities.vision / capabilities.web_search / capabilities.skills
— opt-in capability declarations, not proof the underlying model actually
performs an image/search call correctly. Templates declare
web_search: duckduckgo and vision: inherit (the agent's own provider and
credentials) rather than any per-vendor capability provider. A capability
being wired only means the kernel will attempt the corresponding route;
whether the live call succeeds is a runtime fact, not a manifest fact.test cases passed, vs 10 / 21 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.