Agent审计
- 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
- Bun
- 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: effect-language-service-cc
description: > Use when this capability is needed. Operational guide for @effect/language-service CLI and con…
category: AI 智能
runtime: Bun
---
# effect-language-service-cc 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“Command Decision Table / Interpreting Output / Diagnostic format”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“Command Decision Table / Interpreting Output / Diagnostic format”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 `/tmp` 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“Command Decision Table / Interpreting Output / Diagnostic format”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: effect-language-service-cc
description: > Use when this capability is needed. Operational guide for @effect/language-service CLI and con…
category: AI 智能
source: tomevault-io/skills-registry
---
# effect-language-service-cc
## 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「Command Decision Table / Interpreting Output / Diagnostic format」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 "effect-language-service-cc" {
输入层 -> 用户目标 + 目标文件 + 禁止范围 + 验收标准
上下文层 -> Command Decision Table / Interpreting Output / Diagnostic format
规则层 -> SKILL.md 触发条件 / 执行顺序 / 输出格式
运行层 -> Bun | 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
} Effect Language Service
Operational guide for @effect/language-service CLI and configuration. This skill handles
tooling: running commands, interpreting output, applying fixes, configuring projects.
For architecture decisions about when/how to use Effect, see effect-usage-cc instead.
Command Decision Table
Always run from the project root via the local binary (bunx effect-language-service or
./node_modules/.bin/effect-language-service). Commands like check resolve paths relative
to cwd — running from a subdirectory breaks them.
| Intent | Command | Key flags |
|---|---|---|
| Check for Effect issues | diagnostics --project tsconfig.json |
--format pretty human, --format json parseable |
| Check single file | diagnostics --file <path> |
Faster than full project scan |
| See available fixes | quickfixes --project tsconfig.json |
--code <rule> filter by rule |
| Fix specific issue | quickfixes --file <path> --line <n> --fix <fixName> |
Shows diff to apply |
| Generate from directives | codegen --project tsconfig.json |
--force regenerates all |
| Understand project structure | overview --project tsconfig.json |
--max-symbol-depth 2 (depth 3 adds noise) |
| Analyze layer composition | layerinfo --file <path> --name <Layer> |
Outputs composition order |
| Check TS patch status | check |
Verifies patch version |
| Patch TypeScript for tsc | patch |
--force re-patches |
| Interactive setup | setup |
Wizard for first-time config |
| Configure rule severities | config |
Interactive severity editor |
| CI gate | diagnostics --format github-actions --strict |
Non-zero exit on warnings |
Interpreting Output
Diagnostic format
src/service.ts:42:3 effect(floatingEffect): This Effect is not yielded or assigned
Structure: file:line:col effect(ruleName): message. The ruleName maps directly to the
diagnosticSeverity config key in tsconfig.json and to the inline suppression syntax.
Quickfix format
Output: diagnostic line followed by one or more unified diffs per available fix. Each fix
has a fixName (e.g., floatingEffect_yieldStar) that can be used with --fix to filter.
Severity levels
- error — Must fix. Blocks CI (always non-zero exit)
- warning — Should fix. Blocks CI only with
--strict - message/suggestion — Informational. Never blocks CI even with
--strict(emits::noticein github-actions format, not::error). Only blocks tsc ifincludeSuggestionsInTscenabled
Key Workflows
Fix all issues in a file
# 1. See what's wrong
effect-language-service diagnostics --file src/service.ts --format pretty
# 2. See available fixes with diffs
effect-language-service quickfixes --file src/service.ts
# 3. Filter to a specific rule
effect-language-service quickfixes --file src/service.ts --code floatingEffect
Review the diffs before applying. Correctness fixes (floatingEffect, missingReturnYieldStar) are safe to apply. Style fixes (effectMapVoid, unnecessaryPipe) are preferences — check project conventions.
Onboard a new project
# 1. Interactive setup (patches TS, configures tsconfig)
effect-language-service setup
# 2. Persist patch across installs
# Add to package.json scripts: "prepare": "effect-language-service patch"
# 3. Baseline scan
effect-language-service diagnostics --project tsconfig.json --format pretty
LLM context generation
overview and layerinfo produce text output (no --format json — only diagnostics
supports --format). Both emit ANSI spinner noise that must be stripped.
Clean extraction pattern (use for all text-output commands):
# Strip ANSI codes and spinner lines from any ELS text command
els_clean() {
effect-language-service "$@" 2>/dev/null \
| sed 's/\x1b\[[0-9;]*m//g' \
| grep -v "^Processing file"
}
# Project map: services, layers, yieldable errors
els_clean overview --project tsconfig.json --max-symbol-depth 2
# Specific layer: provides, requires, composition order
els_clean layerinfo --file src/layers.ts --name AppLayer
# Full snapshot: overview + diagnostics + quickfixes in one shot
els_clean overview --project tsconfig.json --max-symbol-depth 2 > /tmp/els-overview.txt
effect-language-service diagnostics --project tsconfig.json --format json > /tmp/els-diag.json
els_clean quickfixes --project tsconfig.json > /tmp/els-fixes.txt
overview groups exports into "Yieldable Errors", "Services", "Layers" with file+line,
type info, and JSDoc. Warning: each symbol appears at every re-export point (barrel
files, index.ts). Deduplicate by source file:line — expect ~40% noise from duplicates
in large projects.
layerinfo outputs a suggested Layer.provide / Layer.provideMerge composition.
Requires a named, exported Layer — anonymous or inline layers are not supported.
Tip: write all layers in Layer.mergeAll(...), run layerinfo, use the suggested order.
Guardrails
Scope your scans — Use
--filewhen working on a single file. Full--projectscans are slow in large codebases and produce noise from files you're not touching.Review quickfix diffs — Correctness fixes are generally safe. Style fixes may conflict with project conventions. Anti-pattern fixes sometimes need manual adjustment.
Monorepo tsconfig — Always specify
--projectexplicitly. Without it, ELS infers the tsconfig which may be wrong in monorepos with multiple configs.codegen --force caution — Regenerates ALL directives in the project. Use
--fileto scope to a single file. Manual edits to generated blocks will be overwritten.Patch persistence — The TS patch modifies
node_modules/typescript/. It disappears afternpm install/bun install. Use"prepare": "effect-language-service patch"to persist it.diagnostics vs tsc — Without patching,
tscdoes NOT run Effect diagnostics. ThediagnosticsCLI command works without patching. The patch is only needed fortscintegration and IDE features.ANSI spinner pollution —
overview,layerinfo, andquickfixesemit ANSI escape codes and "Processing file" spinner lines mixed into stdout. Always pipe through theels_cleanpattern (strip ANSI + filter spinner) when capturing output programmatically.--format is diagnostics-only — Only
diagnosticssupports--format json|text|pretty|github-actions. All other commands (overview,layerinfo,quickfixes,codegen) output unstructured text. Do not waste time trying--formaton them.codegen with no directives — If no files contain
@effect-codegens,codegenthrowsNoFilesToCodegenErrorinstead of exiting cleanly. This is not a real error — the project simply has no codegen directives. Check before running.config/setup file picker — Both
configandsetupare interactive and list ALL.jsonfiles in the project, not just tsconfig files. This is a known UX issue — select the correct tsconfig manually.
Fetching Documentation
This skill covers tooling operations. For current API details or version-specific changes:
| Need | Tool |
|---|---|
| ELS release notes, new rules | mcp__exa__web_search_exa — search Effect-TS/language-service |
| Effect API docs | mcp__Context7__query-docs — resolve effect library |
| Pattern guidance | Load effect-usage-cc skill |
Navigating References
Load references when the command table and workflows above are insufficient.
| File | Answers |
|---|---|
references/cli-commands.md |
Full syntax for each command? All flags and defaults? |
references/diagnostic-rules.md |
What does rule X mean? How to fix it? Default severity? |
references/refactorings.md |
Which IDE refactoring to use? async→Effect variants? |
references/codegen-directives.md |
How to use @effect-codegens? When to annotate/accessors/typeToSchema? |
references/configuration.md |
All tsconfig options? Severity profiles? CI setup? Inline suppression? |
!echo "## ELS Project Context"
!~/.claude/skills/effect-language-service-cc/scripts/probe-els.sh
Source: bengous/agents-skills — 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