chore: squash adjacent standalone commits on nano-as-v1 for merge into main

[mikasenghaas]

Note

Squashed history of feat/nano-as-v1 for the merge into main. Identical content (tree-identical to feat/nano-as-v1), with runs of adjacent standalone (non-PR) commits squashed so the merge carries fewer commits while every individually-merged PR stays its own commit. Merge with a merge commit (not squash) to preserve the PRs. Full original history: #1576.

---

• README - high level overview
• GUIDE - user guide to authoring taskset + harness, cli usage, etc.
• ARCHITECTURE - explanation of framework internals

---

Note

High Risk
Large framework and packaging shift with removed workflows, docs, and environments (bfcl-v3, v1 init path); hub publish and test matrix changes can affect releases and downstream consumers.

Overview
This merge brings in the verifiers v1 stack (import verifiers.v1 as vf): composable taskset × harness configs, swappable runtimes, typed Task/Trace rollouts, and new uv run eval / validate / init entrypoints. Public docs drop the old BYO Harness / prime env init --v1 path in favor of packaged *_v1 environments and TOML under configs/.

Environments move from monolithic load_environment + inline v1 shims to installable packages (e.g. aime24_v1, alphabet_sort_v1, code_golf_v1, color_codeword_v1, example compact harness). Legacy bfcl_v3 and root alphabet_sort_v1.py are removed; v0 alphabet_sort no longer accepts a v1= flag.

CI and release narrow: Semgrep policy and publish-tasksets / publish-harnesses workflows are deleted; ty checks only verifiers; tests drop Python 3.10, add tests/v1, and swap Prime-sandbox tests for v1 runs. Hub auto-publish skips *_v1 dirs and compact until those packages are hub-ready.

^{Reviewed by Cursor Bugbot for commit 71c81bc. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Introduce the verifiers v1 environment framework with tasksets, harnesses, runtimes, and CLI

• Adds a full v1 environment framework under verifiers/v1/ replacing v0-era abstractions with typed Taskset, Harness, Runtime, Rollout, Episode, and Trace components, including MCP-based tool and user simulator servers.
• Introduces concrete harnesses (BashHarness, DefaultHarness, CodexHarness, KimiCodeHarness, MiniSWEAgentHarness, RLMHarness, Terminus2Harness) and runtimes (SubprocessRuntime, DockerRuntime, ModalRuntime, PrimeRuntime) as installable packages.
• Adds a suite of v1 tasksets (AIME24, GSM8K, Math, WikiSearch, Wikispeedia, CodeGolf, AlphabetSort, ColorCodeword, Tau2Bench, SWEBench, R2EGym, ScaleSWE, SWELego, TextArena, Wordle, and others) as independently packaged environments.
• Adds eval, validate, serve, and init CLI entrypoints under verifiers/v1/cli/ with Rich dashboards, resumable runs, dry-run mode, and in-process or worker-pool execution.
• Removes v1 scaffolding support from v0-era utilities (env_utils, scripts/init.py, scripts/eval.py) and strips previously available symbols from verifiers.__init__ and verifiers.types.State.
• Risk: Many symbols previously importable from verifiers and verifiers.v1 are no longer available; verifiers.types now requires the renderers package at import time; State no longer provides runtime/endpoint/tool handle APIs; load_environment raises if the module lacks load_environment; Python 3.10 is no longer supported.

Changes since #1825 opened

• Migrated harnesses from standalone packages/harnesses package into verifiers.v1.harnesses namespace [71c81bc]
• Migrated tasksets from standalone packages/tasksets package into verifiers.v1.tasksets namespace [71c81bc]
• Updated import paths across all environment packages and loader functions to reference consolidated verifiers.v1 namespace [71c81bc]
• Removed standalone package infrastructure for tasksets and harnesses packages [71c81bc]
• Updated dependency specifications to reference consolidated verifiers package with new extras [71c81bc]
• Updated documentation and error messages to reflect consolidated package structure [71c81bc]

^{Macroscope summarized e57c064.}

[xeophon]

lgtm

[macroscopeapp]

Approvability

Verdict: Needs human review

Diff is too large for automated approval analysis. A human reviewer should evaluate this PR.

^{You can customize Macroscope's approvability policy. Learn more.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit e57c064. Configure here.}