Skip to content
Agentmart
Submit a tool

eval-driven-development

Use for nondeterministic Agent rules, Skills, prompts, Hooks, or routing—not deterministic logic, bug fixes with known causes, or deterministic refactors.

Skill by shiyuan17

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/shiyuan17-eval-driven-development

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Use for nondeterministic Agent rules, Skills, prompts, Hooks, or routing—not deterministic logic, bug fixes with known causes, or deterministic refactors.

Use cases: agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/eval-driven-development

What's inside

Skill name
eval-driven-development
Description
Use for nondeterministic Agent rules, Skills, prompts, Hooks, or routing—not deterministic logic, bug fixes with known causes, or deterministic refactors.
License
MIT
Instructions
23 lines in SKILL.md · view on GitHub

Files (3)

  • SKILL.md
  • openai.yaml · agents/
  • metadata.json

SKILL.md (first lines)

# 评测驱动修改 Agent 行为

知识覆盖观察先匹配当前安装的 Rule 和 Skill,并分别保留选择、调用、验证和停止边界。单个 Episode 或缺少调用证据只能是 needs-more-evidence;两个不同 Episode 在相同 request root、完整清单下都确认无匹配后,才可形成 confirmed-uncovered 并进入 owner 评审。该观察不新增完成门禁,也不得保存提示、Session、绝对用户路径、Secrets、命令或消息正文。

同步归属:`docs/rules/eval-driven-development.md`;两者描述同一门禁,修改须同步。

## 执行

1. 用真实输入定义一个会失败的 `Eval-ID`,注明必须行为、禁止行为、风险和确定性断言;capability 变更先确认旧资产能复现失败,旧资产本来就能通过时记为 not-reproduced。
2. 新增或修改 oracle、scoring 或 judge 前先做负控:已知坏样本必须失败,已知好样本不得误判。
3. 修改前冻结 fixture、suite、模型与版本、runner、预算、重复次数、评分标准、rubric 文本、judge 与旧规则/提示指纹的参考结果。
4. 实施最小改动;确定性部分同时运行普通测试。
5. 保持其他实验条件一致,用新规则/提示指纹重跑;被修改资产是实验变量。比较成功率、Token、墙钟、交互、工具调用和错误完成声明,并确认差异超过抖动边界:单轮或只报告成功样本的差异不算改善。
6. critical 必须全部通过且不越过项目配置阈值;reference 更新必须单独审查并显式确认。
7. 确认的真实失败在同一变更内沉淀为 case;长期满分的 case 转入 regression 或替换,不再用作改善信号。

online run 按 `repetitions` 独立运行多轮,输出 `trialSummaries`(`passAt1`/`passAtK`/`passCaretK`/逐轮明细)作为可靠性报告;`passAtK` 是至少一轮通过,`passCaretK` 是全部轮次通过,重复次数少时不构成可靠性结论。稳定性只统计多轮 case 并同时报告覆盖率,当前不新增阈值门禁。offline 为确定性 replay,不输出多轮摘要。

online runner 只传入 model/provider/base URL/reasoning/对应 auth 白名单,并把实际 backend、CLI 版本与 repetitions 纳入 fingerprint。execution case 必须用 `allowedWritePaths` 限定写入;未声明写入或 sandbox/runner 基础设施故障按 fail-closed 处理。普通逐轮诊断不得保存 transcript、命令输出或凭据;一次性合成或公开 Fixture 可在工作区外保存持久化前已脱敏的 ATIF 完整可观察 Trace,原始 Trace 不进 Git、完整 Trace 默认留存不超过 30 天,且任何凭据不得落盘。judge 默认与被测模型分离,复用被测模型自评必须显式声明;语义断言未与人工标注对齐前只能作报告信号。

抖动 case 用 `flaky` 标记并隔离到非门禁观察集,限期修复或替换,隔离项仍进入报告。runner 不可用时报告 degraded,同一 campaign 的 degraded attempt 必须随报告汇总。通用 error item 不计入工具调用,预期安全拒绝与意外失败分开统计。不得自动更新 reference、把缺失运行解释为通过,或只报告双方成功的样本。

From shiyuan17/Vibe-Harness, MIT.

More from shiyuan17/Vibe-Harness

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Oct 14, 2026 (14 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from shiyuan17

Cite this page

eval-driven-development skill. Agent Mart. https://agentmart.dev/skills/shiyuan17-eval-driven-development (last updated 2026-10-24)