Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →lingtai-ai/preset-skill-op-availability-save-gate
Save on the preset editor performs local structural validation only — it never makes a live provider/model network call. Real availability diagnosis is owned by /doctor and actual runtime execution.
Passed 20 of 21 test cases with the skill, and 4 of 21 without it.
and 26 more in the full report
17 fixed4 passed either way1 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 16 test cases the skill fixed. Same model, same prompt, with and without the skill.
Where should a developer look in the documentation hierarchy to understand what happens after a preset is saved, specifically concerning activation and /refresh?
| Checked | Without skill | With skill |
|---|---|---|
| Cites reference/operations/activation-session-refresh/SKILL.md as the relevant operational documentation. | Missed | Passed |
The instructions your model reads when the skill loads.
Evidence: tui/internal/tui/preset_editor.go's commit().
commit() runs the preset's structural Validate() and, if it passes,
saves immediately. There is no live HTTP call, no pending/checking state,
and no per-tuple credential fingerprinting gate on Save — for any
provider family: Codex and Claude Code logins as well as openai- and
anthropic-family API keys. See reference/operations/saved-presets/SKILL.md
for what Validate() checks.
An earlier revision of this editor ran a real live provider/model call on every Save and hard-blocked the save on a non-2xx response (a "real-availability gate"). That gate false-positived broadly — every provider, including working Codex and API-key configurations, could be rejected by transient network/provider conditions unrelated to whether the configuration was actually usable — and was removed (2026-07-23, reported by Jason). It has not been replaced by another probe, whitelist, or retry loop.
/doctor (tui/internal/tui/doctor.go's probeLLM) and real runtime
execution remain the places that diagnose whether a provider/model is
actually reachable. probeLLM checks by provider family: openai lists
{base_url}/models with a Bearer key, anthropic lists
{base_url}/v1/models with x-api-key + anthropic-version (each then makes
one max_tokens=1 generation call to catch empty gateway replies), codex
and claude-code are reported as login-owned without a network call, and any
other provider is reported as unsupported. Save does not duplicate that check.
For structural Validate() and the Load/Save/Delete file contract, see
reference/operations/saved-presets/SKILL.md. For what happens after a
successful save — activation and /refresh — see
test cases passed, vs 4 / 21 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.