Token导航 LogoToken导航

Agent Skill · 研究检索

implement-with-remote-feedback(实施远程反馈)

implement-with-remote-feedback 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

implement-with-remote-feedback 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为执行命令;许可证信息还不明确,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 71,596。

总安装 194周安装 8GitHub Stars 公开资料未说明下载 63核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:implement-with-remote-feedback(实施远程反馈)
来源仓库:https://github.com/xalior/agent-skills
仓库路径:skills/implement-with-remote-feedback
安装命令:
npx skills add https://github.com/xalior/agent-skills --skill implement-with-remote-feedback
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/xalior/agent-skills --skill implement-with-remote-feedback

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

implement-with-remote-feedback 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中根据关键词快速定位候选结果。适用于研究类任务,如查找技术方案、筛选参考资料或匹配开发需求。

通过 npx skills add 命令安装,需结合来源 README 核验具体用法。使用前建议确认权限范围和是否会触发联网或文件操作。适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。

来源仓库内路径:skills/implement-with-remote-feedback

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/xalior/agent-skills --skill implement-with-rem…;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要执行命令,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

implement-with-remote-feedback 怎么安装?

在终端执行 npx skills add https://github.com/xalior/agent-skills --skill implement-with-remote-feedba…。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

implement-with-remote-feedback 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

implement-with-remote-feedback 需要什么权限?

权限级别是执行命令。安装流程涉及命令执行,可能通过 npx skills add https://github.com/xalior/agent-skills --skill implement-with-remote-feedback 联网下载 Skill 或依赖。用户安装前应确认命令来源、仓库内容和执行环境。

implement-with-remote-feedback 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

implement-with-remote-feedback 用的人多吗?

公开数据是累计安装 194 次,近一周安装 8 次,下载 63 次。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

implement-with-remote-feedback 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 需要执行命令,装到常用环境前要先评估影响面。
  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 不能接受 Skill 在你机器上执行命令的人。
  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
implement-with-remote-feedback

实施远程反馈 · 本页

194—执行命令未明确通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Implement with Remote Feedback

Execute a plan document into code, publishing every commit so the remote branch serves as the primary monitoring channel. The plan is your source of truth for WHAT to build and HOW to phase it. You are an implementer, not a designer.

Preflight

Locate the plan. The skill starts from a completed plan. If $ARGUMENTS points to an existing plan doc, use it. Otherwise glob docs/plans/plan_*.md and ask the user to pick one. If no plan exists, STOP and tell the user this skill requires a plan doc as input — do NOT invent scope or design from scratch.

Read the plan in full. Use the Read tool WITHOUT limit/offset. The plan IS your source of truth for scope, phases, and Success Criteria. Do not redefine them. If they need to change, say so and OFFER to switch back to /plan — never switch unilaterally.

Verify the working tree is clean. Run git status. If there are ANY uncommitted or untracked non-ignored changes, STOP and tell the user:

"Working tree is not clean. Please commit or stash your changes before starting."

Verify we are on main/master. If not, warn the user and ask whether to continue from the current branch or switch to main first.

Pull latest. git pull to ensure we're up to date.

Confirm the branch name with the user. The default is <type>/<slug>, where <slug> matches the plan's slug. Valid types:

  • feature/ — new functionality
  • bugfix/ — fixing a defect
  • spike/ — exploratory / research / prototype
  • refactor/ — restructuring without behaviour change
  • docs/ — documentation only
  • chore/ — maintenance, deps, tooling

Present the single default and ask the user to confirm or override. Do not enumerate alternatives beyond the type list above. Once confirmed, create and publish the branch:

git checkout -b <type>/<slug>
git push -u origin <type>/<slug>

The Doc

Create the implementation tracker at docs/plan_<slug>_implementation.md with this skeleton:

# Implementation: <title from plan>

**Status:** In Progress
**Branch:** `<type>/<slug>`
**Plan:** `<path to plan_<slug>.md>`

## Tasks

- [ ] Phase 1: <name>
- [ ] Phase 2: <name>

## Progress Log

## Decisions & Notes

## Blockers

## Commits

The tracker is a living document — update it continuously. It tells the story of the implementation to anyone reading the git log. The file on disk plus the remote branch are the persistence layer.

Commit AND push the tracker immediately with a message like chore: init implementation tracker for <slug>.

The Work

Execute the plan's phases in order. The stance is skepticism — if a Success Criterion isn't verifiable (Automated = a command to run; Manual = a specific thing to observe), it isn't done. NEVER mark a phase complete on vibes.

  • FOLLOW THE PLAN. Execute its phases in the order they're written. NEVER batch across phases, NEVER skip ahead, NEVER silently merge two phases.
  • HONOR THE APPROACH. If the plan specifies vertical slices (the default), each phase cuts end-to-end through the stack — e.g. DB → model → server → api → client lib → frontend, or whichever layers the feature touches. NEVER complete one layer across all features when the plan calls for slices. If the plan specifies something else, follow it as written.
  • INVESTIGATE BEFORE ASKING. Before questioning the user, read the plan, read referenced files in full, and look at the current code. Spawn research sub-agents in parallel when broad coverage is needed. Then ask only what investigation can't answer. NEVER ask the user about things you could have looked up.
  • READ REFERENCED MATERIALS IN FULL. The plan. Files named in Key Discoveries. Related code the plan references. Use the Read tool WITHOUT limit/offset. NEVER skim. NEVER summarise-and-move-on.
  • VERIFY, DO NOT ADOPT. Claims the user makes mid-implementation — about constraints, existing behaviour, intent in the plan — get verified before they change direction. Corrections to your own statements ALSO get verified. Spawn a sub-agent to verify where possible.
  • WAIT FOR ALL SUB-TASKS TO COMPLETE before acting on their findings. Be patient with sub-agents and vocal about them: say what you have spawned, and speak up when each one returns.
  • Update the tracker FIRST, then do the work. Mark the current task in progress, append a progress log entry with a timestamp. Then write the code.
  • Prefer Makefile targets for verification (make test, make lint, make -C <subproject> check). The plan's Success Criteria should name them; run what the plan names. If a needed target is missing, extend the Makefile as part of this phase.
  • Run ALL Automated Success Criteria before marking a phase complete. Every checkbox. No selective verification.
  • Pause for Manual Success Criteria. When a phase has Manual criteria, STOP after Automated checks pass and tell the user exactly what needs observing. Do NOT proceed to the next phase until they confirm.
  • Commit early, commit often, in small logical units. One reason per commit. If you touched 5 files for 3 reasons, that's 3 commits. Prefixes:

- feat: — new functionality - fix: — bug fix - wip: — partial work (always with context: wip: partial auth middleware, never just wip) - docs: — documentation - test: — tests only - refactor: — restructure without behaviour change - chore: — tooling, deps, maintenance

  • PUSH AFTER EVERY COMMIT. NO EXCEPTIONS. The remote branch IS the monitoring channel. git add <specific-files> git commit -m "<type>: <description>" git push
  • Update the tracker after each commit — tick off tasks, append progress, note decisions or blockers. Commit and push the tracker update too.
  • If blocked, record the blocker in the tracker, commit, push, then ask the user. Do NOT spin on a blocker silently.
  • When all phases are complete and all Success Criteria met, set the tracker's Status to Complete, make a final commit and push, and tell the user the branch is ready. Offer to open the PR.

Monitoring

Others can watch progress via:

git log --oneline origin/<type>/<slug>
git diff main..origin/<type>/<slug>

Or by reading docs/plan_<slug>_implementation.md on the remote branch.

Never

  • NEVER skip pushing. Every commit must be pushed immediately. The remote branch is the monitoring channel — a local-only commit defeats the entire purpose of this skill.
  • NEVER force push. History is sacred in this workflow.
  • NEVER amend pushed commits. Make a new commit instead.
  • Never redefine scope, phases, or Success Criteria. Those come from the plan. If they need to change, OFFER to switch back to /plan — never switch unilaterally, and never redefine in-place.
  • Never mark a phase complete without running its Automated criteria and obtaining user confirmation on its Manual criteria.
  • Never batch multiple phases into one commit or one chunk of work.
  • Never commit with vague messages. wip alone is not a commit message.
  • Never lead the user with unsolicited alternatives.

平台分布

Codex

35.78%
按下载量换算23

Claude

29.09%
按下载量换算18

Cursor

19.44%
按下载量换算12

Gemini CLI

9.2%
按下载量换算6

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

执行命令

安装流程涉及命令执行,可能通过 npx skills add https://github.com/xalior/agent-skills --skill implement-with-remote-feedback 联网下载 Skill 或依赖。用户安装前应确认命令来源、仓库内容和执行环境。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills