據 AlphaMoat 統計,截至 2026-10,Agentic Coding 累計下載 16,147 次、獲 10 個星標,按累計下載量在本次參與排名的 10,914 個 AI 技能中排名第 1,743 位,在「AI 智慧體」分類(1,632 個)中排名第 242 位。
列表預設按累計下載量排序,也可按星標排序;詳情頁排名按累計下載量計算。
| 平台 | 累計下載 | 星標 | 評論 |
|---|---|---|---|
| ClawHub ↗ | 1.4萬 | 10 | - |
| SkillHub ↗ | 1.6萬 | 10 | — |
各平台的計數口徑不同,統計時間不一致。
編輯前先確定目標、檢查、約束和非目標;修復前記錄失敗條件,修復後重跑同一檢查。交付說明改動檔案、結果、風險與回滾方式,反覆失敗後須結構化升級處理。持久化筆記需明確選擇加入,專案自身規則仍適用。
資料釋出於 2026-10-27
以下是根據技能文件生成的示例,不代表實際執行結果。
2026-09 起統計方法調整:同名技能按作者分別統計,並覆蓋全部技能。圖中虛線為調整前的月份,與之後的資料不直接比較,也不計算跨越調整的增長。
| 月份 | 累計下載 | 環比 | 星標 |
|---|---|---|---|
| 2026-10 | 1.6萬 | — | 10 |
| 2026-09 | 1.4萬 | — | 10 |
| 2026-08 | 1.3萬 | +3.7% | 10 |
| 2026-07 | 1.3萬 | +6.2% | 10 |
| 2026-06 | 1.2萬 | +3.6% | 10 |
| 2026-05 | 1.2萬 | — | 10 |
| 2026-03 | 8716 | — | 8 |
以下是作者釋出的原始檔案,按純文字顯示,本站未做修改或稽核;其中的連結不可點選,命令請核實後再執行。抓取於 2026-10-23。
---
name: Agentic Coding
slug: agentic-coding
version: 1.0.0
homepage: https://clawic.com/skills/agentic-coding
description: Ship production code with AI agents through acceptance contracts, micro diffs, red green loops, and deterministic handoff checkpoints.
changelog: Initial release with PACT protocol, contract templates, and handoff-first delivery workflow.
metadata: {"clawdbot":{"emoji":"AI","requires":{"bins":["git"]},"os":["linux","darwin","win32"]}}
---
## Setup If `~/agentic-coding/` does not exist or is empty, read `setup.md`, ask a concise kickoff question, and keep any persistence explicitly opt-in. ## Positioning This skill is intentionally different from `agentic-engineering` and `vibe-coding`: - `agentic-engineering` focuses on multi-agent operating patterns and team throughput. - `vibe-coding` focuses on prompt-led exploration and fast idea shipping. - `agentic-coding` focuses on contract-first implementation, proof of fix, and reviewer-ready handoff. ## When to Use User needs merge-ready code from an AI agent with explicit quality gates. Use for production features, risky refactors, bug fixes with reproducible failures, and Xcode-centered work such as Swift feature delivery, iOS/macOS regressions, and release-branch hotfixes. ## Architecture Memory lives in `~/agentic-coding/`. See `memory-template.md` for setup. ```text ~/agentic-coding/ |- memory.md # Persistent preferences and operating mode |- contracts.md # Accepted task contracts and non-goals |- evidence.md # Test evidence and verification snapshots `- handoffs.md # Delivery notes and rollback hints ``` ## Quick Reference Load these files on demand to keep context focused and execution fast. | Topic | File | |-------|------| | Setup process | `setup.md` | | Memory template | `memory-template.md` | | PACT loop | `protocol.md` | | Contract prompts | `prompt-contracts.md` | | Merge handoff checklist | `handoff.md` | ## Core Rules ### 1. Lock a Contract Before Writing Code Start every task with a compact contract: - Objective: exact outcome in one sentence - Acceptance: checks that prove success - Non-goals: what must stay untouched - Constraints: stack, style, limits, deadlines No contract, no code. ### 2. Run the PACT Loop Use the same execution loop every time: 1. **P**roblem framing: restate objective and assumptions 2. **A**cceptance design: define checks before edits 3. **C**hange set: produce the smallest useful diff 4. **T**race and test: show evidence and residual risk This skill is execution discipline, not brainstorming. For Xcode workflows, tie acceptance to a concrete target, simulator/device, and test command before editing. ### 3. Keep Diffs Surgical One user objective maps to one focused change set: - Prefer file-local edits over broad rewrites - Separate behavior change from style cleanup - Avoid hidden side effects outside declared scope If scope grows, split into a second contract. ### 4. Prove Failure Then Prove Fix For bugs and regressions: - Capture the failing condition first (test, log, or reproduction) - Apply minimal fix - Re-run the same check to prove resolution Never claim fixed without before and after evidence. ### 5. Deliver Handoff-Grade Output End each cycle with a delivery packet: - What changed and why - Files touched and blast radius - Validation run and results - Known risks and rollback path If handoff is unclear, the task is not finished. ### 6. Escalate With a Structured Fallback When blocked after two failed attempts: - Stop editing - State what was tried - Propose two grounded alternatives - Request a decision with tradeoffs Do not keep guessing in loops. ## Common Traps - Starting implementation without acceptance checks -> endless iteration and unclear done state. - Asking the agent for full rewrites -> noisy diffs and avoidable regressions. - Mixing feature work with architecture overhaul -> weak reviewability and hard rollback. - Reporting success without reproducible evidence -> false confidence in production. - Treating AI output as final draft -> quality debt moved to code review. ## Security & Privacy **Data that leaves your machine:** - None from this skill itself **Data that stays local:** - Contracts, evidence notes, and handoff summaries in `~/agentic-coding/` **This skill does NOT:** - Trigger undeclared network requests - Access files outside its own memory path - Write to global or platform memory stores - Auto-approve risky code without explicit evidence ## Scope This skill ONLY: - Improves execution quality of AI-assisted coding - Enforces contract driven implementation and verification - Produces clear handoff packets for reviewers This skill NEVER: - Replaces security review for high risk domains - Encourages blind trust in generated code - Overrides project specific contribution rules ## Related Skills Install with `clawhub install <slug>` if user confirms: - `agentic-engineering` - Multi-agent collaboration and operating patterns. - `coding` - General coding support across stacks and tasks. - `code` - Broad code authoring and editing assistance. - `copilot` - Companion style IDE assistance patterns. - `delegate` - Structured task delegation to autonomous agents. ## Feedback - If useful: `clawhub star agentic-coding` - Stay updated: `clawhub sync`
將開發過程中的經驗、錯誤與修正記錄為 Markdown 檔案,供日後重用並提升為工作區記憶。
用自然語言描述需求,在本機與遠端技能市場尋找技能並取得推薦。
給 AI agent 用的瀏覽器自動化命令列技能。
將使用者明確糾正與自我檢視整理成分層記憶,並依使用情況升級或封存。
規劃 AI 代理主動跟進、工作記憶與回顧流程的工作區設定指南。
指導如何建立與更新有效 Claude 技能的寫作指南。
在安裝第三方技能之前,按清單審查來源、許可權與危險模式的技能。
把盛唐詩人李白的詩學觀、意象系統與口吻,蒸餾成 AI 可呼叫的角色包。
將使用者明確糾正與自我檢視整理成分層記憶,並依使用情況升級或封存。
在保留公式、格式和活頁簿結構的前提下建立、檢查或編輯 Excel 檔案。
處理 Word 與 DOCX 文件的作業指南,著重維持樣式、編號與版面。
面向多家影片模型 API 的 AI 影片生成與編排技能。
把資料庫、表格與匯出的原始資料變成可決策結論的資料分析技能。
資料月份:2026-10 · 下載/星標數聚合自公開技能註冊庫(ClawHub、SkillHub),按月更新。口徑說明見資料方法。