Agent转换
- 作者仓库星标 0
- 作者更新于 实时读取
- 作者仓库 skills-registry
- 领域
- AI 智能
- 兼容 Agent
-
- Claude Code
- Cursor
- Cline
- Codex
- Windsurf
- Gemini CLI
- +20
- 信任分
- 88 / 100 · 社区维护
- 作者 / 版本 / 许可
- @tomevault-io · 未声明 license
- 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
- macOS · Linux · Windows
- 底层运行要求
- Node.js
- 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: adevsync
description: Sync constitution to CLAUDE.md, AGENTS.md, and other agent files declared in manifest.yaml. Run…
category: AI 智能
runtime: Node.js
---
# adevsync 输出预览
## PART A: 任务判断
- 适用问题:提示词、Agent 工作流、模型评估或自动化推理。
- 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“Provider Detection / Process / Dry-Run Mode”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提示词、Agent 工作流、模型评估或自动化推理,并标出输入、限制和预期结果。
- **02** 执行计划:优先按“Provider Detection / Process / Dry-Run Mode”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。 原文出现了 `/adev` 这类斜杠命令;如果你的 Agent 支持命令触发,优先用命令开场,再补充目标和边界。
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“Provider Detection / Process / Dry-Run Mode”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: adevsync
description: Sync constitution to CLAUDE.md, AGENTS.md, and other agent files declared in manifest.yaml. Run…
category: AI 智能
source: tomevault-io/skills-registry
---
# adevsync
## 什么时候使用
- 把 AI / Agent方向的常用动作沉淀成 Agent 可调用的技能 适合处理AI Agent、提示词、模型评估与自动化推理,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…
- 面向提示词、Agent 工作流、模型评估或自动化推理,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「Provider Detection / Process / Dry-Run Mode」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 "adevsync" {
输入层 -> 用户目标 + 目标文件 + 禁止范围 + 验收标准
上下文层 -> Provider Detection / Process / Dry-Run Mode
规则层 -> SKILL.md 触发条件 / 执行顺序 / 输出格式
运行层 -> Node.js | 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
} Sync Constitution to Agent Files
Reads .context-index/constitution.md and generates tool-specific agent files based on manifest.yaml sync targets.
Provider Detection
When syncing, detect which AI coding assistant is running:
- Claude Code:
CLAUDE.md(primary) - OpenCode:
AGENTS.md(primary) - Cursor:
.cursorrules - GitHub Copilot:
.github/copilot-instructions.md
If multiple providers are used, sync all enabled targets from the manifest.
Process
Read source files:
.context-index/constitution.md(required).context-index/manifest.yaml(required, for sync targets).context-index/platform-context.yaml(optional, for tech stack summary)
For each sync target in manifest:
Claude format (
CLAUDE.md)<!-- Synced from .context-index/constitution.md by adev. Do not edit above the User Additions line. --> [Full constitution content] ## Context Index This project uses the Agentic Development Framework (adev). - Constitution: `.context-index/constitution.md` - Manifest: `.context-index/manifest.yaml` - Platform: [summary from platform-context.yaml] - Available skills: /adev:brainstorm, /adev:specify, /adev:review-specs, /adev:plan, /adev:implement, /adev:validate, /adev:debug, /adev:issues, /adev:hygiene ## Task Management (conditional) <!-- BEGIN TASK MANAGEMENT --> [Read `tasks.backend` from manifest.yaml. If configured: - Include the Task Management section from constitution.md. - If tasks.backend is "beads", include br command reference. - If tasks.backend is "file", include markdown table reference. - If constitution has no Task Management section, generate from the default content matching the configured backend. If tasks.backend is NOT configured in the manifest, omit this entire section (no empty block, no placeholder).] <!-- END TASK MANAGEMENT --> ## Learned Lessons (conditional — see step 3) # User Additions <!-- Content below is preserved across syncs. Add Claude-specific instructions here. --> [preserved content from previous CLAUDE.md below this marker]OpenCode / Generic agents format (
AGENTS.md)<!-- Synced from .context-index/constitution.md by adev. --> [Full constitution content] ## Project Context This project uses the Agentic Development Framework (adev). - Constitution: `.context-index/constitution.md` - Manifest: `.context-index/manifest.yaml` - Platform: [summary from platform-context.yaml] - Available skills: /adev:brainstorm, /adev:specify, /adev:review-specs, /adev:plan, /adev:implement, /adev:validate, /adev:debug, /adev:issues, /adev:hygiene ## Task Management (conditional) <!-- BEGIN TASK MANAGEMENT --> [Same logic as Claude format: include if tasks.backend is configured, omit otherwise.] <!-- END TASK MANAGEMENT -->Copilot format (
.github/copilot-instructions.md)Principles and coding standards only. Omit Context Routing (Copilot has limited file navigation).
Cursor format (
.cursorrules)Full constitution content. Convert Context Routing pointers to Cursor-compatible references where applicable.
Inject Learned Lessons (conditional):
Read high-confidence heuristics via inline Node.js using
retrieveHeuristicsandrenderHeuristicfromlib/heuristics.mjs. This step is non-blocking: ifretrieveHeuristicsthrows, log a warning to stderr and proceed without the section.import { retrieveHeuristics } from 'lib/heuristics.mjs'; const heuristics = await retrieveHeuristics(projectRoot, '_global'); const highOnly = heuristics.filter(h => h.confidence === 'high');- If no
high-confidence entries exist, skip the section entirely. If a stale## Learned Lessonsblock is already present in the target file, remove it (replace from the heading to the next##heading or EOF, whichever comes first). - If entries exist, group by scope alphabetically. The
_globalscope sorts last. - Render each entry as:
- <title> (<scope>) — <pattern truncated to 80 chars> - Placement in CLAUDE.md and AGENTS.md: Place the
## Learned Lessonssection heading immediately before the# User Additionsmarker. - Placement in .cursorrules and copilot-instructions.md (
.github/copilot-instructions.md): Append the section at the end of the file. - On re-sync: Detect an existing
## Learned Lessonsheading and remove the old block (from the heading to the next##or EOF), then write the fresh replacement in the correct position.
- If no
Preserve User Additions:
- Look for
# User Additionsmarker in existing target file - If found, preserve everything below the marker
- If marker missing, append the marker with empty section
- If target file does not exist, create it fresh
- Look for
Report: List which files were updated and their line counts.
Dry-Run Mode
/adev:sync --dry-run — Show what would be generated without writing files. Print the diff for each target.
When to Run
- After
/adev:init(automatic) - After editing
.context-index/constitution.md(suggested by sync-trigger hook) - After editing
.context-index/manifest.yaml(manual) - When agent files seem stale (
/adev:hygienewill detect this)
Context Routing referent
The constitution's Context Routing table includes a row Lifecycle state | .context-index/lifecycle-state/ — this points at the per-spec JSONL event logs managed by lib/lifecycle-state.mjs. When syncing the constitution into agent files, propagate this row verbatim; do not rewrite it to refer to the legacy build-state/ directory (which has been renamed by one-shot-migration-tool.spec.md).
API reference
Heuristic injection (Step 3):
retrieveHeuristics(projectRoot, scope)andrenderHeuristic(entry)from<ADEV_ROOT>/lib/heuristics.mjs— read high-confidence heuristics for the## Learned Lessonsblock.
Manifest:
loadManifest(projectRoot)from<ADEV_ROOT>/lib/manifest.mjs— parses.context-index/manifest.yaml. Use when reading sync targets,tasks.backend, andproject.adev_version.
Source: agentic-development/adev-plugin — distributed by TomeVault.
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