Skip to content

GeoTask Core v0.1.0 Public Preview

Pre-release
Pre-release

Choose a tag to compare

View all tags
@stpku stpku released this 27 Jul 10:20
v0.1.0-public-preview
f8cdb86

GeoTask Core v0.1.0 Public Preview

  • Release date: 2026-07-27
  • Git tag: v0.1.0-public-preview
  • Package version: 0.1.0
  • Document Schema: 1.0

中文说明 | English notes

中文发布说明

GeoTask Core v0.1.0 Public Preview是GeoTask首个可引用、可下载、可运行的公共预览版本。

GeoTask面向AI智能体提供一套可验证时空任务协议：模型负责理解任务并提出候选对象、断言和动作；GeoTask Core负责解析、规范化、验证引用，并通过本地确定性算子复算结果。模型生成的答案不会被直接当作已经验证的结论。

本次发布包含

  • 六类Canonical对象：point、polyline、rect、time_interval、altitude_interval、feature_collection；
  • 六个本地确定性算子：
    • distance_2d
    • line_intersects_rect
    • point_to_line_distance_2d
    • rect_contains_point
    • time_overlap
    • altitude_overlap
  • YAML任务解析、Canonical IR、结构验证和确定性执行；
  • 结构化结果、ClaimStatus、来源与Assurance元数据；
  • 模型输出Normalizer与本地Verifier；
  • geotask validate、geotask run、geotask inspect等CLI能力；
  • GeoTask Language and Execution Specification 1.0；
  • Draft 2020-12 JSON Schema；
  • GT01—GT13渐进式应用案例，覆盖空间关系、时空组合、证据治理、机器人协同、无人机能源余量和车辆安全包络；
  • 中文项目门户、白皮书、Quickstart、Cookbook、贡献指南和社区模板；
  • Python 3.10、3.11、3.12和3.13持续集成；
  • 公共导出白名单、敏感信息扫描和架构边界检查。

快速开始

git clone https://github.com/stpku/GeoTask.git
cd GeoTask
pip install -e ".[dev]"

geotask validate examples/core/v1_minimal_distance.yaml
geotask run examples/core/v1_minimal_distance.yaml
pytest -q

发布资产

GitHub Release提供：

  • Python wheel：geotask_core-0.1.0-py3-none-any.whl；
  • Source distribution：geotask_core-0.1.0.tar.gz；
  • GitHub自动生成的源代码归档；
  • 本发布说明和固定版本Tag。

验证状态

  • 公共仓库测试：336项通过；
  • Python 3.10—3.13 CI：全部通过；
  • GitHub Pages门户与GT01—GT13：全部上线；
  • Secret Scanning与Push Protection：已启用；
  • 公共导出扫描：未发现密钥、内部路径或二进制泄露。

当前定位

本版本是Public Preview，而不是稳定的1.0产品版本。当前重点是验证协议表达、确定性执行、模型结果比较、公开案例和开发者体验。未来路线见ROADMAP.md。

兼容性

项目早期名称为STIR。为降低迁移成本，本版本仍保留部分stir字段、CLI和函数兼容入口，但新项目应统一使用GeoTask命名。详见MIGRATION.md。

引用

研究论文、技术报告和软件项目可使用仓库根目录的CITATION.cff生成推荐引用格式。

English Release Notes

GeoTask Core v0.1.0 Public Preview is the first citable, downloadable, and runnable public preview of GeoTask.

GeoTask provides a verifiable spatiotemporal task protocol for AI agents. Models propose objects, assertions, explanations, and candidate actions. GeoTask Core parses and canonicalizes the task, validates references and operator contracts, and recomputes supported claims with deterministic local operators.

Included in this release

  • six canonical object types: point, polyline, rect, time_interval, altitude_interval, and feature_collection;
  • six deterministic local operators: distance_2d, line_intersects_rect, point_to_line_distance_2d, rect_contains_point, time_overlap, and altitude_overlap;
  • YAML parsing, Canonical IR, structural validation, and deterministic execution;
  • structured results, claim status, source, and assurance metadata;
  • model-output normalization and local verification;
  • CLI commands including geotask validate, geotask run, and geotask inspect;
  • GeoTask Language and Execution Specification 1.0;
  • a Draft 2020-12 JSON Schema;
  • GT01–GT13 progressive cases covering spatial relations, spatiotemporal composition, evidence governance, robot coordination, UAV energy reserve, and vehicle clearance envelopes;
  • a project portal, white paper, Quickstart, Cookbook, contributor guides, and community templates;
  • CI on Python 3.10 through 3.13;
  • public-export allowlisting, secret scanning, and architecture-boundary checks.

Quickstart

git clone https://github.com/stpku/GeoTask.git
cd GeoTask
pip install -e ".[dev]"

geotask validate examples/core/v1_minimal_distance.yaml
geotask run examples/core/v1_minimal_distance.yaml
pytest -q

Release assets

The GitHub Release includes the Python wheel, source distribution, GitHub source archives, these release notes, and an immutable version tag.

Status

This is a Public Preview rather than a stable 1.0 product release. The current focus is protocol representation, deterministic execution, model-result comparison, public application cases, and developer experience. See ROADMAP.md for planned public directions.

Assets 4
Loading