Token导航 LogoToken导航TokenDH.com
效率操作浏览器clawhub未标认证来源可访问clear审计提醒

jumpstartjumpstart 效率

Agent Skill

jumpstart 用于补充效率相关能力,适合在 OpenClaw 中需要让 Agent 承接效率相关任务时使用。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。

总安装

3,585

周安装

154

GitHub Stars

公开资料未说明

下载量

1,257
OpenClaw

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

MIT-0

最后核验

2026-05-01

来源状态

来源可访问

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:jumpstart(jumpstart 效率)
来源仓库:https://github.com/kzac313/jumpstart
安装命令:
openclaw skills install jumpstart
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 OpenClaw 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。

ClawHubOpenClaw
openclaw skills install jumpstart

简介

jumpstart 用于补充效率相关能力,适合在 OpenClaw 中需要让 Agent 承接效率相关任务时使用。

  • 适用于复杂多会话编码项目的代理工具设置与管理,支持长期运行任务。
  • 通过 clawhub 安装,结合来源仓库和 README 核验具体用法。
  • 安装前建议确认权限范围、维护状态及是否触发联网、命令执行或文件读写。
  • 使用时需参考原始 SKILL.md 了解详细功能与限制条件。

SKILL.md

name
jumpstart
description
Sets up and manages a long-running agent harness for complex, multi-session coding projects. Use this skill whenever a user invokes jumpstart, jumpsession, jumploop, or jumpfree, or when they want to build something substantial that won't fit in a single context window — web apps, large codebases, multi-feature products — or when they ask about autonomous coding agents, multi-session agents, agent loops, or keeping progress across sessions. Also trigger when the user asks how to structure an agent to avoid losing context, how to make an agent work incrementally, or how to prevent an agent from declaring victory too early.

Jumpstart: Long-Running Agent Harness

A skill for setting up and running a two-command agent harness that enables sustained, incremental progress across many context windows — based on Anthropic's *Effective Harnesses for Long-Running Agents*.


Commands

CommandWhen to useWhat it does
jumpstartOnce, at project startReads design files, scaffolds the environment, generates feature_list.json
jumpsessionEvery subsequent sessionOrients from progress.txt, picks one feature, implements, commits, logs
jumploop --nBatch implementation sessionsLike jumpsession but loops through up to n features without stopping
jumpfreeAd-hoc or exploratory sessionsOrients from progress.txt, implements based on user input, commits, logs

Input: Design Files

Before running jumpstart, the project directory should contain these three files. If jumpstart is called and any of these files are missing, create blank versions of them:

FileContents
GENERAL_DESIGN.mdHigh-level product vision, goals, non-goals, target users
UX_DESIGN.mdUser flows, interface descriptions, interaction patterns
ENG_DESIGN.mdTechnical architecture, components, data models, APIs, constraints

jumpstart derives feature_list.json primarily from ENG_DESIGN.md, cross-referencing the other two for completeness.


Environment Files Created by jumpstart

FilePurpose
init.shStarts the dev server and runs a basic smoke test
feature_list.jsonExhaustive feature list derived from ENG_DESIGN.md, all initially not_started
progress.txtRunning log of what each session accomplished
Initial git commitBaseline snapshot the agent can always revert to

jumpstart Prompt

Use for the very first session only.
You are initializing a long-running coding project. Your job is NOT to build the app — it's to read the design documents and set up everything the next agent will need.

STEP 1 — Read the design documents:
- Read GENERAL_DESIGN.md for product vision and goals
- Read UX_DESIGN.md for user flows and interface patterns
- Read ENG_DESIGN.md for technical architecture and component breakdown
- IMPORTANT: If any of these three design files do not exist, create them as blank files before proceeding.

STEP 2 — Create init.sh:
A script that starts the development server and runs a basic end-to-end smoke test
(e.g., verify the server starts, the app loads, and one core interaction completes).
Make it runnable: chmod +x init.sh

STEP 3 — Create feature_list.json:
Derive a comprehensive list of features from ENG_DESIGN.md, supplemented by UX_DESIGN.md.
Each feature must have:
  - "id": unique string, e.g. "feat-001"
  - "category": functional | ui | performance | security | infrastructure
  - "description": plain English, specific and testable
  - "steps": array of steps a human tester would follow to verify end-to-end
  - "status": "not_started"  ← always not_started to begin
  - "notes": ""  ← empty string, for agents to leave observations

Be exhaustive. 50–200+ features for a real app. Vague features are useless.
Good: "User can submit the login form with valid credentials and be redirected to the dashboard"
Bad: "Login works"

STEP 3b — Insert human verification checkpoints:
After generating the feature list, insert checkpoint entries at logical milestone
boundaries — places where a human should pause and verify the app before moving on.
Checkpoints must be placed at transitions that represent meaningful integration
points, for example:
  - After all core UI components are implemented
  - After basic navigation/routing is working
  - After backend API is functional
  - After frontend-backend integration is connected
  - After authentication/authorization is wired up
  - Before performance or polish work begins

Each checkpoint must have:
  - "id": unique string, e.g. "checkpoint-001"
  - "type": "checkpoint"  ← distinguishes from regular features
  - "description": what milestone was just reached, e.g. "Core UI shell complete"
  - "verify": array of short items a human should check at this point
  - "status": "not_started"
  - "notes": ""

Place 3–6 checkpoints across a typical project. They must appear in the correct
position in the feature list (after the features they gate). Do not cluster them —
spread them at natural breakpoints.

STEP 4 — Create progress.txt:
Start with a header: "PROJECT: [project name]" and today's date.
Add a single entry: "Session 0 — Initialization complete. feature_list.json generated with [N] features and [M] checkpoints."

STEP 5 — Initial git commit:
git add -A && git commit -m "init: project scaffold, feature list, and environment setup"

RULES:
- Do not implement any features.
- Do not change any feature's status from not_started.
- Do not modify the design files.

jumpsession Prompt

Use for every session after the first.
You are a coding agent resuming work on an ongoing project.

START OF SESSION — do these steps in order:
1. Run `pwd` to confirm your working directory.
2. Read `progress.txt` — focus on the most recent session entry.
3. Run `git log --oneline -20` to review recent commits.
4. Run `init.sh` to start the dev server and confirm there is no broken build.
   If it IS broken, fix the regression before starting new work.
5. Read `feature_list.json`. Choose the highest-priority feature with status "not_started".
   Prefer features that unblock other features.
   If the next item is a checkpoint ("type": "checkpoint"), handle it first (see below).

DURING YOUR SESSION:
- Work on EXACTLY ONE feature per session. Stop after it is resolved.
- When you begin a feature, update its status to "in_progress".
- Implement, then verify the feature works. Prefer inspecting the DOM or reading server
  output over opening a browser URL — do not let an inability to open a URL block you.
  Browser automation tools (e.g., Puppeteer) may be used when readily available, but are
  not required.
- When verified, set status to "complete" and commit immediately:
  `git add -A && git commit -m "feat([id]): [feature name] — [brief summary]"`
- If you hit a hard blocker (missing dependency, design ambiguity, unresolvable bug),
  set status to "blocked", commit the current state, and document the reason in "notes".
- NEVER remove or edit a feature's description or steps — only update status and notes.
- NEVER mark a feature "complete" without having verified it.

CHECKPOINTS:
- If the next item in the feature list is a checkpoint ("type": "checkpoint"),
  do NOT skip it. Present the checkpoint's "verify" list to the user and ask them
  to confirm everything looks good before proceeding.
- Set the checkpoint's status to "awaiting_review" and stop the session.
- The user will mark it "complete" (or provide feedback) before the next session.

END OF SESSION — after committing the feature, do these final steps:
1. Append to progress.txt:
   - Date and session number
   - Feature worked on and its final status
   - Any bugs found and whether they were fixed
   - Blockers or open questions
   - Recommended next feature for the following session
2. Stop. Do not begin a second feature.

RULES:
- It is unacceptable to mark a feature complete without verifying it.
- It is unacceptable to edit or remove feature descriptions or steps.
- It is unacceptable to declare the project done unless every feature is "complete" or "deferred".
- Always leave the codebase in a mergeable state — no broken builds, no half-implemented features.
- Never end a session with a feature left in "in_progress" — resolve to "complete" or "blocked".
- Each feature must get its own git commit before the session ends.
- One feature per session — do not start a second feature after the first is resolved.

jumploop Prompt

Use when you want to implement multiple features in a single session without manual intervention. Invoke as jumploop --n where n is the number of features to complete.
You are a coding agent resuming work on an ongoing project. You will implement
multiple features in this session — up to [N] features total.

START OF SESSION — do these steps in order:
1. Run `pwd` to confirm your working directory.
2. Read `progress.txt` — focus on the most recent session entry.
3. Run `git log --oneline -20` to review recent commits.
4. Run `init.sh` to start the dev server and confirm there is no broken build.
   If it IS broken, fix the regression before starting new work.
5. Read `feature_list.json`. Choose the highest-priority feature with status "not_started".
   Prefer features that unblock other features.

FEATURE LOOP — repeat until you have completed [N] features or no "not_started" features remain:
1. Set the chosen feature's status to "in_progress".
2. Plan the implementation. Do NOT ask for user approval — proceed immediately.
3. Implement the feature.
4. Verify the feature works. Prefer inspecting the DOM or reading server output
   over opening a browser URL.
5. When verified, set status to "complete" and commit immediately:
   `git add -A && git commit -m "feat([id]): [feature name] — [brief summary]"`
6. If you hit a hard blocker, set status to "blocked", commit, document the reason
   in "notes", and move on to the next feature — do not stop the loop.
7. Pick the next highest-priority "not_started" feature and repeat.

CHECKPOINTS:
- If the next item is a checkpoint ("type": "checkpoint"), the loop MUST stop
  regardless of how many features remain in the budget.
- Present the checkpoint's "verify" list to the user and ask them to confirm
  everything looks good.
- Set the checkpoint's status to "awaiting_review" and end the session.
- The user will mark it "complete" (or provide feedback) before the next session.
- Checkpoints do NOT count toward the [N] feature budget.

RULES:
- Plans always proceed — never pause to ask for user review or approval.
- Complete up to [N] features, but stop early if all remaining features are
  "complete", "blocked", or "deferred", or if a checkpoint is reached.
- NEVER remove or edit a feature's description or steps — only update status and notes.
- NEVER mark a feature "complete" without having verified it.
- Always leave the codebase in a mergeable state after each commit.
- Never leave a feature in "in_progress" — resolve to "complete" or "blocked"
  before moving to the next feature.
- Each feature must get its own git commit.

END OF SESSION — after all features are resolved, do these final steps:
1. Append to progress.txt:
   - Date and session number
   - For each feature worked on: feature ID, description, and final status
   - Any bugs found and whether they were fixed
   - Blockers or open questions
   - Total features completed / attempted
   - If stopped at a checkpoint: checkpoint ID and verify list
   - Recommended next feature for the following session
2. Stop.

jumpfree Prompt

Use for ad-hoc sessions where the objective is defined by immediate user input rather than the pre-planned feature list.
You are a coding agent resuming work on an ongoing project.

START OF SESSION — do these steps in order:
1. Run `pwd` to confirm your working directory.
2. Read `progress.txt` — focus on the most recent session entry.
3. Run `git log --oneline -20` to review recent commits.
4. Run `init.sh` to start the dev server and confirm there is no broken build.
   If it IS broken, fix the regression before starting new work.
5. Review the user's specific request for this session.

DURING YOUR SESSION:
- Focus solely on fulfilling the user's specific request.
- Implement, then verify the work. Prefer inspecting the DOM or reading server
  output over opening a browser URL — do not let an inability to open a URL block you.
- When verified, commit immediately:
  `git add -A && git commit -m "feat(jumpfree): [brief summary of work done]"`

END OF SESSION — after committing the work, do these final steps:
1. Append to progress.txt:
   - Date and session identifier (e.g. jumpfree)
   - Ad-hoc task worked on and its final status
   - Any bugs found and whether they were fixed
   - Blockers or open questions
   - Recommended next feature for the following session (optional)
2. Stop. Do not begin additional unsolicited work.

RULES:
- Always leave the codebase in a mergeable state — no broken builds.
- The work must get its own git commit before the session ends.

feature_list.json Format

[
  {
    "id": "feat-001",
    "category": "functional",
    "description": "User can open the app and see a welcome screen with a prompt to log in",
    "steps": [
      "Navigate to http://localhost:3000",
      "Verify the page loads without console errors",
      "Verify a welcome message is visible",
      "Verify a login button or link is present"
    ],
    "status": "not_started",
    "notes": ""
  },
  {
    "id": "checkpoint-001",
    "type": "checkpoint",
    "description": "Core UI shell complete — all base components and layouts are implemented",
    "verify": [
      "App loads without errors",
      "All pages are reachable via navigation",
      "Components render correctly across viewport sizes",
      "Visual design matches UX_DESIGN.md intent"
    ],
    "status": "not_started",
    "notes": ""
  }
]

Entry Types

FieldFeature entryCheckpoint entry
idfeat-NNNcheckpoint-NNN
type*(absent or "feature")*"checkpoint"
descriptionWhat the feature doesWhat milestone was just reached
steps / verifysteps — verification stepsverify — items for a human to check
statusSee status tableSee status table
notesAgent observationsHuman feedback

Status Values

StatusMeaning
not_startedDefault. No work has begun.
in_progressActively being worked on in the current session. Never persist this across sessions.
completeImplemented and verified end-to-end. Do not revisit unless a regression is found.
blockedCannot proceed — dependency missing, design unclear, or unresolvable bug. Document reason in notes.
deferredIntentionally postponed by agreement. Document reason in notes.
awaiting_reviewCheckpoint only. Human review requested — agent must stop and wait.

Key rules:

  • JSON only (not Markdown) — models are less likely to accidentally overwrite JSON
  • Agents may only change status and notes — never description or steps/verify
  • in_progress must never persist across sessions — end each session with complete or blocked
  • Checkpoints gate progress — agents must not skip past an awaiting_review or not_started checkpoint

progress.txt Format

PROJECT: [Name]
Started: [Date]

---

Session 0 — [Date]
Initialization complete. feature_list.json generated with 87 features.

---

Session 1 — [Date]
Worked on: feat-001 (User welcome screen)
Status: complete
Summary: Implemented the landing page with login CTA. Verified via Puppeteer — page loads, welcome text visible, login button present.
Bugs fixed: None.
Blockers: None.
Next session: feat-002 (Login form submission)

---

Session Startup Checklist (Quick Reference)

[ ] pwd
[ ] Read progress.txt (latest entry)
[ ] git log --oneline -20
[ ] Run init.sh
[ ] Smoke test — is the app healthy?
[ ] (jumpsession only) Read feature_list.json → pick 1 "not_started" feature
[ ] (jumpsession only) Set feature to "in_progress"
[ ] (jumploop only) Read feature_list.json → begin feature loop (up to N features)
[ ] (jumploop only) Plans always auto-proceed — no user approval needed
[ ] (jumpfree only) Review user's ad-hoc request
[ ] Implement → verify (DOM/server output preferred over opening URL) → set to "complete" or "blocked"
[ ] git add -A && git commit (per task, immediately after resolving)
[ ] Append to progress.txt
[ ] (jumpsession/jumpfree) Stop — do not begin a second feature/task
[ ] (jumploop) Stop after N features completed or no features remain

适合场景

01

OpenClaw 用户查找和安装 Skill 时

02

用户想查找某类 Agent Skill 时

03

需要根据任务场景推荐可安装能力包时

04

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

OpenClaw

74.89%
按下载量换算941

安全审计

VirusTotal

可疑

ClawScan

通过

Static analysis

通过

权限和风险

操作浏览器

该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

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。来源安全扫描存在 warning/failed 结果,不能写成本站确认安全。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills