测试诊断
- 作者仓库星标 367
- 许可证 NOASSERTION
- 作者更新于 实时读取
- 作者仓库 mngr
- 领域
- 通用
- 兼容 Agent
-
- Claude Code
- Cursor
- Cline
- Codex
- Windsurf
- Gemini CLI
- +20
- 信任分
- 94 / 100 · 已通过审计
- 作者 / 版本 / 许可
- @imbue-ai · NOASSERTION
- Token 消耗评级
- 低消耗
- 接入复杂程度
- 即装即用
- 是否需要外部 API Key
- 不需要
- 兼容的系统
- 未声明(默认跨平台)
- 底层运行要求
- Python
- 文件与系统权限
-
- 只读
- 允许写入 / 修改
- 网络行为
- 仅限本地
- 安装命令数
- 26 条
档案由构建时根据 SKILL.md 与安装命令自动衍生,可能与作者实际意图存在差异。
需要注意: 未限定 allowed-tools,默认拥有全部工具权限。
---
name: fix-something
description: Fix a random FIXME in the codebase (in the given scope, use "." for the whole codebase) Your tas…
category: 通用
runtime: Python
---
# fix-something 输出预览
## PART A: 任务判断
- 适用问题:通用任务拆解、检查和交付。
- 输入要求:目标材料、限制条件、期望输出和验收方式。
- 证据边界:围绕“先判断是否适合 / 作者设计意图 / 作者的方法与取舍”读取原文规则,不把推断写成作者承诺。
## PART B: 执行结果
- **01** 任务判断:确认你的需求是否属于通用任务拆解、检查和交付,并标出输入、限制和预期结果。
- **02** 执行计划:优先按“先判断是否适合 / 作者设计意图 / 作者的方法与取舍”拆成步骤,说明每一步会读取什么、修改什么、产出什么。
- **03** 交付结果:给出可复制的命令、文件改动、检查清单或内容草稿,并说明如何继续迭代。
- **04** 风险边界:结合 读取文件、写入/修改文件、主要在本地完成、通常不需要额外 API Key 给出执行前确认项。
## Running Rules
- 读取文件、写入/修改文件;主要在本地完成;通常不需要额外 API Key。
- 先小样例验证,再放大到真实任务。
- 交付时同时给结果、检查口径和下一步迭代建议。 原文没有稳定的斜杠命令要求。安装验证后通常全局生效,直接在对话里点名这个 Skill 并描述任务即可。
告诉 Agent 目标文件或材料、期望结果、不可改范围、是否允许联网或执行命令。本 Skill 的权限画像是:读取文件、写入/修改文件。
先用一个小任务确认它会围绕“先判断是否适合 / 作者设计意图 / 作者的方法与取舍”工作;涉及文件或命令时,先看 diff、日志、预览或测试结果。
检查最终产物是否包含明确结果、必要证据和下一步动作;如果输出泛泛而谈,就补充输入、边界和验收标准后重跑。
---
name: fix-something
description: Fix a random FIXME in the codebase (in the given scope, use "." for the whole codebase) Your tas…
category: 通用
source: imbue-ai/mngr
---
# fix-something
## 什么时候使用
- 把「修复」相关任务沉淀成 Agent 可调用的技能 适合处理通用任务拆解、检查、交付和复盘,核心价值是把输入、判断、执行、验证和交付边界固定下来,避免 Agent 泛泛回答。 把任务拆成可执行、可检查、可继续迭代的步骤;通常不需要额外…
- 面向通用任务拆解、检查和交付,优先处理能明确输入、步骤和验收标准的工作。
## 需要提供什么
- 目标材料、目录范围、期望结果和不可改动内容。
- 是否允许联网、执行命令、读写文件或调用外部服务。
## 执行规则
- 围绕「先判断是否适合 / 作者设计意图 / 作者的方法与取舍」组织步骤,不把推断写成作者事实。
- 读取文件、写入/修改文件;主要在本地完成;通常不需要额外 API Key。
- 先跑小样例,确认结果可检查后再扩大任务范围。
## 输出要求
- 给出最终产物、关键证据、验证方式和下一步动作。
- 信息不足时标记 unknown,不编造命令、平台或依赖。 作者原文负责流程事实;仓库文件负责来源和命令;流狐只补充适用场景、限制和质量判断。
skill "fix-something" {
输入层 -> 用户目标 + 目标文件 + 禁止范围 + 验收标准
上下文层 -> 先判断是否适合 / 作者设计意图 / 作者的方法与取舍
规则层 -> SKILL.md 触发条件 / 执行顺序 / 输出格式
运行层 -> Python | 读取文件、写入/修改文件 | 主要在本地完成
安全层 -> 通常不需要额外 API Key + 小任务验证 + diff / 日志复核
输出层 -> 可复制结果 + 检查清单 + 下一步迭代
} Your task is to make ONE improvement to the codebase by fixing a random thing about the codebase (eg, normally a FIXME).
In order to make this process easier for you, our FIXMEs are specially formatted:
# FIXME(priority)[attempts=N]: (description)
# (optional additional context)
where description is a short description of what needs to be fixed, and N is the number of prior attempts made to fix it (if any).
If there have been no prior attempts, the [attempts=N] part may be omitted.
The priority is simply an integer, with 0 being the highest priority. priority may or may not be present.
If not present, assume priority=3
Before doing any of the below, the first step is to check that the tests are passing by running "uv run pytest"
If the tests are not ALL passing (including all linters, ratchets, type checks, etc), then this is the thing for you to fix. In particular, you must follow this process:
- revert the changes from the previous commit by calling "git revert --no-commit HEAD" (so that the changed are in your current working tree, but not yet committed)
- update the FIXME that was attempted in the previous commit to increment the attempts count by 1 (if there were no prior attempts, add "[attempts=1]", if there were some prior attempts, increment the number by 1). Note that it should then look something like this:
# FIXME[attempts=1]: (original description)or# FIXME0[attempts=2]: (original description), ie, the[attempts=N]part goes before the ":" - extend the optional additional context of the FIXME with a brief note about why you were unable to fix it
- commit all of this.
- proceed directly to "# The Final Step" below.
If the tests start out passing, then you must select and fix a random FIXME by following this process:
- Simply run this bash command, and it will give you a random FIXME line: "./scripts/random_fixme.sh $1". If no lines are returned, then there are no more remaining FIXMEs, so use your "think-of-something-to-fix" skill to come up with something else to fix instead.
- Find that FIXME and be sure to read the surrounding context (the optional additional context lines below the FIXME line may be important). This is the task you will be working on.
- Go gather all the context for the library that contains that FIXME (per instructions in CLAUDE.md).
- Think carefully about how best to fix that FIXME
- Implement the fix
- Get all the tests passing (use "uv run pytest" to run them, and be sure to fix any issues, no matter how small)
Once the tests are passing, be sure that you have removed ONLY that FIXME, then commit your changes.
If you were unable to fix the issue and get all the tests passing, do the following instead:
- Revert any changes you made while attempting to fix the issue
- Update the FIXME to increment the attempts count by 1 (if there were no prior attempts, add "[attempts=1]", if there were some prior attempts, increment the number by 1). Note that it should then look something like this:
# FIXME[attempts=1]: (original description)or# FIXME0[attempts=2]: (original description), ie, the[attempts=N]part goes before the ":" - Extend the optional additional context with a brief note about why you were unable to fix it this time.
- Commit this updated FIXME (be sure that nothing else is being changed!)
The Final Step
Finally, as the very last thing you do (after committing, regardless of whether you succeeded or not), run this command: "touch $2" (to indicate that you have completed your fix task).
先判断是否适合
作者设计意图
作者的方法与取舍
边界和复核