← 返回 Skills 市场
cirokk

GSD-Claw

作者 Ciro Borges · GitHub ↗ · v1.0.1 · MIT-0
cross-platform ✓ 安全检测通过
137
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install gsdclaw
功能描述
Spec-first execution workflow for OpenClaw that turns large, vague, or messy project requests into phased, verifiable implementation plans. Helps build, resc...
使用说明 (SKILL.md)

GSD-Claw

Run medium and large project work without letting scope, structure, or context drift.

Quick Start

Use this workflow when a request is too large, vague, messy, or failure-prone to execute in one pass.

Default sequence:

  1. map current state
  2. rewrite request into a build spec
  3. define phases
  4. set acceptance criteria
  5. execute in narrow batches
  6. verify after each meaningful step
  7. report real status without pretending everything is done

Security Notes

  • This skill is instruction-only.
  • It does not require credentials, external APIs, or special binaries.
  • It should not claim integrations are real unless they were actually implemented and verified.
  • If a project contains secrets, private URLs, or credentials, keep them out of reusable/public skill content.

Core Workflow

1. Map the current state

  • Inspect the existing project before proposing major changes.
  • Identify stack, entrypoints, broken areas, constraints, and likely risks.
  • Prefer short factual notes over speculation.

2. Convert the request into an execution spec

  • Rewrite the user request into a short internal build spec.
  • Separate:
    • objective
    • required outcomes
    • quality bar
    • exclusions / unknowns
    • dependencies / blockers
  • If the request is broad, define a practical v1 that still honors the user’s intent.

3. Define phases

  • Break the work into small phases with visible outputs.
  • Each phase must be independently verifiable.
  • Prefer this shape when applicable:
    1. stabilize current project
    2. architecture / structure cleanup
    3. UX / UI redesign
    4. core features
    5. integrations / configuration
    6. validation / polish

4. Set acceptance criteria before implementation

  • Define what “done” means in observable terms.
  • Prefer criteria such as:
    • page renders
    • button works
    • form persists
    • config saves
    • layout is responsive
    • build passes
    • no obvious runtime errors

5. Execute in narrow batches

  • Avoid giant rewrites unless they are clearly safer.
  • Prefer a short loop:
    • edit
    • verify
    • continue
  • When files are large or write operations are fragile, work in smaller modules and validate after each change.

6. Verify aggressively

  • After each meaningful step, verify with the strongest available method:
    • file inspection
    • build/run
    • browser/manual rendering check
    • test command
    • diff/output review
  • Never assume success from writing alone.

7. Report like a builder

  • Summarize:
    • what changed
    • what was verified
    • what remains
    • current blockers
  • Do not claim completion if major paths are still broken.

Operating Rules

  • Prefer one canonical implementation path. Remove ambiguity before multiplying files.
  • Do not leave duplicate competing files unless there is a clear migration reason.
  • If the codebase gets fragmented, stop and consolidate before adding more features.
  • If the request includes redesign, prioritize information architecture and usability before visual polish.
  • If the request includes integrations (APIs, providers, messaging, AI), separate:
    • UI/config capture
    • config persistence
    • connection test
    • real runtime integration
  • Clearly distinguish simulated integration from real integration.

Anti-Context-Drift Rules

  • Re-anchor on the user goal before each major phase.
  • Prefer short execution plans over long narrative plans.
  • Keep a single source of truth for current state.
  • If implementation starts drifting, rewrite the project status as:
    • done
    • in progress
    • blocked
    • next
  • If a previous attempt produced broken or truncated files, treat that as technical debt and fix it before adding scope.

Rescue / Salvage Mode

When the existing project is already in a bad state:

  1. identify the canonical entrypoint
  2. identify duplicated or broken artifacts
  3. choose what to keep
  4. remove or ignore dead paths
  5. restore a minimal working baseline
  6. only then continue feature work

Prefer restoring one working path over preserving every partial attempt.

UI/UX Mode

When improving UI/UX:

  • fix structure before styling
  • improve hierarchy, spacing, labels, navigation, and flows first
  • ensure actions are obvious and feedback is visible
  • make primary actions dominant and secondary actions quieter
  • ensure empty states, loading states, and error states exist where relevant

Configuration-Heavy Products

When the app includes provider setup pages, model the feature in this order:

  1. provider selector
  2. credential fields
  3. help/instructions
  4. save state
  5. connection test
  6. runtime usage

Never present a configuration screen as fully functional if it only stores fields locally. Make the distinction explicit in implementation notes and validation.

Completion Standard

A task is only finished when:

  • the main path works end-to-end, or
  • the remaining gap is explicitly limited to external dependency/blocker work

If blocked, provide the smallest next unblock step.

File Reference

Read these when needed:

  • references/execution-template.md — convert vague requests into a practical build spec
  • references/acceptance-checklist.md — completion criteria and verification prompts
安全使用建议
This skill is an instruction-only workflow for managing and executing software work; it appears coherent and doesn't request secrets or install code. Before using it: (1) be aware the agent will inspect your repository files and run verifications — avoid exposing secrets or private credentials in the repo or prompt; (2) supervise first runs (confirm each proposed phase and acceptance criteria) so the agent doesn't unintentionally modify important files; (3) treat integrations/configuration steps as potentially simulated unless you explicitly provide and verify real connection tests; and (4) verify the GitHub homepage/source if you want provenance (metadata lists a homepage but source/owner in the registry is not a known vendor). If you need the agent to perform actions that require credentials or external services, grant them deliberately and minimally.
功能分析
Type: OpenClaw Skill Name: gsdclaw Version: 1.0.1 The gsd-claw skill is a purely instructional framework designed to guide an AI agent through structured project management and execution workflows. It contains no executable code, requires no special permissions or environment variables, and focuses on best practices for planning, verification, and maintaining code quality. The instructions in SKILL.md and the reference templates (acceptance-checklist.md, execution-template.md) promote transparency and rigorous verification rather than any form of data exfiltration or unauthorized access.
能力评估
Purpose & Capability
The name and description (spec-first execution/workflow for managing projects) match the SKILL.md content: phased plans, acceptance criteria, verification, and rescue modes. The skill does not request unrelated binaries, environment variables, or external services.
Instruction Scope
The instructions focus on inspecting a codebase, producing phased specs, running targeted edits, and verifying results. These actions are appropriate for the stated purpose. Note: the workflow expects the agent to inspect project files, run builds/tests, and verify outputs — which is expected but means the agent will access repository files and any secrets stored there if present. The SKILL.md explicitly warns to keep secrets out of reusable skill content.
Install Mechanism
No install spec and no included code files — instruction-only. Nothing will be written to disk or downloaded as part of installing this skill, minimizing install-time risk.
Credentials
No required environment variables, primary credentials, or config paths are declared. This aligns with a workflow/guide that only instructs the agent how to operate against a project repository or user-provided context.
Persistence & Privilege
The skill does not request always:true or other elevated persistent privileges. It is user-invocable and can be invoked autonomously (platform default), which is appropriate for a productivity workflow.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install gsdclaw
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /gsdclaw 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.1
- Added metadata, license (MIT-0), category (productivity), emoji, and homepage to skill header for discoverability. - Improved and lengthened the description with clear usage scenarios. - Added Quick Start, Security Notes, and File Reference sections for easier onboarding and safety guidance. - Clarified core workflow steps and formatted them as actionable instructions. - No implementation/code changes; documentation and metadata update only.
v1.0.0
Initial release of GSD-Claw – a spec-first execution workflow for OpenClaw projects. - Provides a detailed step-by-step workflow for turning vague requests into phased, verifiable plans. - Emphasizes mapping current state, breaking down tasks, setting clear acceptance criteria, and verifying outcomes. - Includes dedicated guidance for rescue/salvage scenarios, UI/UX improvements, and configuration-heavy products. - Enforces strong rules to prevent scope, structure, and context drift in medium to large technical projects. - Outlines clear completion standards and introduces reference materials for spec writing and acceptance checks.
元数据
Slug gsdclaw
版本 1.0.1
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 2
常见问题

GSD-Claw 是什么?

Spec-first execution workflow for OpenClaw that turns large, vague, or messy project requests into phased, verifiable implementation plans. Helps build, resc... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 137 次。

如何安装 GSD-Claw?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install gsdclaw」即可一键安装,无需额外配置。

GSD-Claw 是免费的吗?

是的,GSD-Claw 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

GSD-Claw 支持哪些平台?

GSD-Claw 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 GSD-Claw?

由 Ciro Borges(@cirokk)开发并维护,当前版本 v1.0.1。

💬 留言讨论