Skip to content

Releases: yifanes/agent-harness-rs

v0.2.6

17 Jun 06:05

Choose a tag to compare

Added

  • Added web_fetch as a built-in read-only tool across mock, local, sandbox, and e2b runtimes.
  • Added hosted-tool abstraction via HostedTool.
  • Added Anthropic native hosted web_search support through web_search_20250305.
  • Added AgentLoopHarness::with_hosted_tools(...) and AgentLoopHarness::with_web_search().
  • Added silent-stop detection for empty/whitespace-only model turns with no tool calls.

Changed

  • PlanApproval now treats web_fetch as read-only.
  • OpenAI-compatible Chat Completions now fails fast when hosted tools are requested; OpenAI web search is reserved for a future Responses API client.
  • Anthropic streaming explicitly ignores provider-hosted web-search blocks.

Fixed

  • Prevented fetched/cached model-limit entries from reducing known offline fallback limits.

v0.2.2

11 Jun 11:11

Choose a tag to compare

feat(skills): progressive-disclosure skill catalogue

新增 skills 模块:向系统提示注入紧凑的 name/description/path 技能目录,而非把每个 SKILL.md 正文塞进 prompt。

  • 技能文件位置通过 SkillSource trait 抽象(默认 LocalSkillSource 走本地 FS,可由 sandbox 实现覆盖)。
  • 模型按需用已有的 read 工具拉取完整指令、用 bash 运行脚本,无需专门的 execute-skill 工具。
  • 导出:SkillsManager / SkillLoader / SkillPromptRenderer / SkillSource / LocalSkillSource / SkillMetadata / parse_skill_frontmatter

v0.2.1

11 Jun 06:26

Choose a tag to compare

借鉴 MiMoCode 工具层机制的三项改造:

新增 / 改进

  • 错误感知截断bounded_preview / clip_overflow 在截断超长工具输出时按 head 70% / tail 30% 保留尾部报错段,不再丢失错误信息;sandbox 与 local 运行时共享同一实现,消除重复的 head-only 逻辑。
  • 统一执行包装层 BoundedToolRuntime<R>:在 AgentLoopHarness::new 中包裹所有运行时——tracing span、幂等的 schema 修复(ToolRuntime::repair_invocation 作为单一事实源)、轻量校验(失败给教学式错误)、以及对未自限工具(MCP / 自定义)的兜底输出上限。
  • 成功静默 / 失败教学edit 成功路径不再向模型回传 repair 字段;非法入参时追加由 schema 合成的 Expected shape 示例(validate_against_schema / example_for_schema)。

修复

  • e2b unary 解码路径中既有的 clippy::never_loop

测试 182 全绿;--all-features clippy 无 error。