Imported from joy7758/dcell-external-evaluation (
AGENTS.md). Install upstream withnpx skills add joy7758/dcell-external-evaluation. Copyright stays with the author.
DCell External Evaluation Agent Entry
Read STATUS.json, UPSTREAM_LOCKS.json, and README.md before acting.
This repository contains research-only adapters for pinned third-party benchmarks. Upstream data, scenarios, runners, scorers, metrics, prompts, failure rules, and leaderboard algorithms are immutable inputs.
Truth boundaries:
ADAPTER_NE_BENCHMARK=true
UPSTREAM_METRIC_NE_DCELL_METRIC=true
EXTERNAL_RESULT_NE_EXTERNAL_VALIDATION=true
SUBMISSION_DRAFT_NE_SUBMISSION=true
INTERNAL_RESULT_REMAINS_INCONCLUSIVE=true
Never store or print API keys. Never use test tasks to build learning artifacts. The repository was published once as a sanitized single-root research snapshot. Do not create tags, releases, benchmark submissions, issues, pull requests, or further public mutations without a separate, explicit owner authorization.