feat(bench): EnterpriseOps-Gym adapter — deployable SQL state-checker (the gate's non-coding middle-band domain)#157
Merged
Conversation
The gate's ideal NON-coding domain: a deployable programmatic verifier (the agent's tool-call transcript is replayed against a freshly-seeded gym MCP server, then the task's own SQL state-checks run via /api/sql-runner) — NOT an answer- oracle. Graded per-verifier (score = passes/total = the bench's verifier_level_pass_rate; resolved = all-pass = overall_success_rate). Covers the enterprise-ops user story (itsm/hr/csm/calendar/email/drive/teams). Real adapter on the existing pattern: loadTasks pulls the HF dataset (ServiceNow-AI/EnterpriseOps-Gym) with a committed offline fixtures fallback; judge invokes the live-env driver and FAILS LOUD with the exact docker pull/run/ seed step when the gym server is unreachable — never a fabricated score. Registered in adapters.ts (one line). goldArtifact undefined (the oracle is the seeded DB state, not a portable transcript — documented, not faked). Note: dataset card says Apache-2.0; the paper lists CC-BY-NC-SA — confirm before redistribution. Verified: bench tsc 0, fixture test 5/5, repo lint clean.
drewstone
added a commit
that referenced
this pull request
Jun 6, 2026
Cuts the 58-commit backlog on main into a published release. Headline surface: - runToolLoop / streamToolLoop — bounded turn-level tool-dispatch loop (#137) - RSI agent tree: recursive Agent.act, Supervisor keystone, runProgram, the adaptive-driver channel (#139/#151/#165) - optimization API collapsed onto agent-eval selfImprove; the runtime keeps the CODE-surface ImprovementDriver you pass as driver (#172) - deployable benchmark adapters: AppWorld, commit0, aec-bench, EnterpriseOps-Gym; runBenchmarks over one ADAPTERS registry (#153/#156/#157) - agent-eval floor raised to >=0.83.0 (#175)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Wires EnterpriseOps-Gym (ServiceNow-AI) as a real
BenchmarkAdapterthrough the unifier. It's the gate's ideal domain: a deployable programmatic verifier — the agent's tool-call transcript is replayed against a freshly-seeded gym MCP server, then the task's own SQL state-checks run via /api/sql-runner (NOT an answer-oracle, NOT an LLM-judge). Graded per-verifier (score = passes/total; resolved = all-pass), covering the enterprise-ops user story. loadTasks pulls the HF dataset with a committed offline fixture; judge fails loud with the exact docker pull/run/seed step when the gym server is unreachable — never a fabricated score. One line in adapters.ts. Verified: bench tsc 0, fixture test 5/5, lint clean. Note: dataset card Apache-2.0 vs paper CC-BY-NC-SA — confirm before redistribution.