v1.61.0.0 feat: add Antigravity host and benchmark support

[KiDDarn]

What changed

• add Antigravity (Agy) as a declarative gstack host, including setup and plugin installation support
• add the Agy model-benchmark provider and document the host and benchmark workflows
• harden Codex, Gemini, Claude, and Agy benchmark adapters around stdin, readiness checks, token accounting, and provider errors
• classify temporary model-capacity failures separately from authentication, rate-limit, timeout, and unknown failures
• fix related setup generation, gbrain installation detection, session import, and secret-sink test behavior

Why

This makes gstack usable from Antigravity and allows subscription-backed CLI providers to be compared through one benchmark runner. The follow-up hardening prevents dry runs from claiming unusable legacy Gemini paths are ready, avoids nested Codex waiting on inherited stdin, removes inflated Codex token overhead, and reports provider saturation accurately.

Impact

Users can install and use gstack from Agy, run provider benchmarks without metered API integration, and receive actionable failure categories when a selected model is temporarily at capacity. Existing hosts keep their current behavior.

Root causes addressed

• provider probes checked binary/auth presence without validating the actual supported CLI path
• nested Codex inherited stdin and interpreted it as additional input
• benchmark token accounting included wrapper/context overhead rather than the intended payload
• provider exceptions were classified from stderr alone, while some CLIs expose capacity details through the exception message
• a fixed four-part version string triggered the pre-push public-IP heuristic

Validation

• npm test (pass)
• focused benchmark/provider/setup suite: 456 passed, 12 skipped, 0 failed
• git diff --check (pass)
• gstack redaction scan over added lines and new files: 0 HIGH, 0 MEDIUM, 0 LOW, 0 WARN

[trunk-io]

Merging to main in this repository is managed by Trunk.

• To merge this pull request, check the box to the left or comment /trunk merge below.

After your PR is submitted to the merge queue, this comment will be automatically updated with its status. If the PR fails, failure details will also be posted here

[time-attack]

@16francej — recommend closing this PR because #2134 supersedes this host-integration approach. This branch deletes its newly generated plugin skill source and hardcodes a mismatched manifest version. Credit #2244 for the plugin and setup intent, while #2134 should fix source ownership, derive the version, and verify that installed skills are nonzero.

[KiDDarn]

This broad PR is being decomposed instead of rebased wholesale:

• fix: honor skipped Claude skills in validation #2282 preserves the independent Claude skill-check validator fix.
• fix: harden benchmark provider readiness and accounting #2283 preserves the generic Codex/Gemini/Claude benchmark hardening.
• The remaining unique Agy setup/runtime safeguards and reference commits have been documented on feat(hosts): add Antigravity (agy) declarative config and setup registry #2134.

I am leaving this PR in draft temporarily as the reference implementation for Agy setup/runtime work. Once those pieces are absorbed into #2134 (or a focused follow-up), this PR should be closed as superseded.