Skip to content

feat: model and host adapters (release evaluation PR 3) - #40

Closed
noodlemind wants to merge 1 commit into
feat/terminal-bench-adapterfrom
feat/eval-host-adapters
Closed

feat: model and host adapters (release evaluation PR 3)#40
noodlemind wants to merge 1 commit into
feat/terminal-bench-adapterfrom
feat/eval-host-adapters

Conversation

@noodlemind

Copy link
Copy Markdown
Owner

Summary

PR 3 of the release-evaluation plan: the model and host adapters that the release orchestrator (PR 4) schedules.

  • openrouter-kimi.mjs — the controlled API experiment on the pinned Kimi K2.7 Code profile. Credential validation fails closed: no OPENROUTER_API_KEY, no driver, no spend. Gates the release after calibration.
  • ollama-gemma.mjs — the local capability floor: free, credential-less, informational gate, with the Docker-container endpoint rewrite (host.docker.internal) from the plan.
  • codex-subscription.mjs / claude-subscription.mjs — the manual frontier A/B contracts (scheduled rotation). Each defines operator run instructions (fresh sandboxes, record the exact resolved model, never mix numerically with the OpenRouter result), transcript preservation to disk, and host-report normalization where every unavailable or malformed telemetry field is recorded as null — never silently estimated.
  • copilot-smoke.mjs / grok-smoke.mjs — the host-compatibility checklist (install/hydration, agent discovery, skill activation, hook behavior, completion+verifier). Evaluation fails closed: unreported checklist items count as failures.

Test coverage

9 new deterministic tests (eval-hosts.test.mjs): pinned-profile wiring, fail-closed credential validation, Docker endpoint rewrite, the all-null subscription telemetry template, normalization of malformed host reports, transcript preservation, and fail-closed smoke evaluation.

  • Full suite: 734/734 pass

Stack

Based on #39 (PR 2: Terminal-Bench adapter), which is based on #38 (PR 1). PR 4 adds release orchestration and reporting.

Controlled Kimi/OpenRouter host with fail-closed credentials, local
Gemma/Ollama host with Docker endpoint rewrite, Codex and Claude
subscription A/B contracts with transcript preservation and null-not-
estimated telemetry normalization, and Copilot/Grok compatibility smoke
checklists with fail-closed evaluation.
@coderabbitai

coderabbitai Bot commented Jul 31, 2026

Copy link
Copy Markdown

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 14e3fc82-c8e2-4946-bd69-d7a4b69b5dd0

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

  • 🔍 Trigger review

Comment @coderabbitai help to get the list of available commands.

@noodlemind

Copy link
Copy Markdown
Owner Author

Consolidated into #38 per review direction — all four workstreams ship as one PR.

@noodlemind noodlemind closed this Jul 31, 2026
@noodlemind
noodlemind deleted the feat/eval-host-adapters branch July 31, 2026 01:29
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant