yzmw123/dify-workflow-dsl-skill

최근 기록된 소스 활동: 2026년 10월 30일 07:12

SkillsMP 카탈로그 업데이트: 2026년 11월 12일 12:52

수집된 skills: 1

GitHub 스타: 104
GitHub 포크: 7

AI 어시스턴트로 설치

이 프롬프트를 복사해 사용 중인 AI 어시스턴트에 입력하세요.

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/yzmw123/dify-workflow-dsl-skill.

저장소 README

README 원본 파일: README.md

README 원본 보기

README 전체 펼치기

# Dify Workflow DSL Skill > Create, repair, review, migrate, and validate Dify App DSL with an AI coding > agent. [中文说明](./README_CN.md) · [4 Dify 1.18 examples](./examples/dify-1.18.0) · [10 Dify 1.16 examples](./examples/dify-1.16.0) · [1.18 evaluation report](./references/dify-1.18-evaluation.md) > **Targets Dify 1.18 by default, with Dify 1.15/1.16 compatibility.** Both > 1.16 and 1.18 use DSL `0.7.0`; select the Dify release with `--dify-version`. > [1.18 compatibility](./references/dify-1.18-target.md) · > [Current verification](./references/dify-1.18-evaluation.md) ## What's New — 2026-10-30 - **Dify 1.18 support with legacy profiles.** Select 1.18, 1.16, or 1.15 with `--dify-version`; DSL 0.7.0 defaults to 1.18 and DSL 0.6.0 to 1.15. - **Current Agent and knowledge schemas.** Validate Agent system outputs and routes, workspace skills, KnowledgeFS bindings, `knowledge-retrieval-v2`, and portable Site settings. Four new examples demonstrate these features. - **Self-contained dependency setup.** `run_validator.py` prepares a reusable virtual environment; `install.sh --setup` can prepare it during installation. Offline setup, JSON output, and the original CLI remain supported. Compatibility and installation fixes: - Distinguish legacy plugin Agents using `version: "2"` from portable Dify Agents, and accept missing-resource `file_id: null` placeholders under 1.18. - Reject stale route edges after Agent routing is disabled. - Restore pip in an interrupted virtual environment and prevent forced installation from overwriting its own source files. - Report malformed node types and app modes without interrupting batch validation; use Dify's production resolver for source-based node checks. **Verified:** 54 local regression tests passed. All 14 maintained examples and 2 additional compatibility cases passed Dify 1.18.0 source checks. All 14 examples were persisted through Dify's real database import path; two produced three expected warnings about workspace resources that need rebinding. Model execution and UI save/re-export round trips remain workspace-level checks. See the [verification record](./references/dify-1.18-evaluation.md) for evidence, commands, and limitations. ## 🎯 Version Strategy > [!IMPORTANT] > **Dify 1.18.x → App DSL `"0.7.0"`** > > This is the default for new workflows and includes portable **Agent Apps** and > **Agent v2 workflow nodes**. | Target workspace | DSL version | Policy | | --- | --- | --- | | **Dify 1.18.x** | **`"0.7.0"`** | Default for new generation | | Dify 1.16.x | `"0.7.0"` | Legacy field and output rules | | Dify 1.15.x | `"0.6.0"` | Compatibility generation and validation | If Dify reports that the DSL version is incompatible, first check the target workspace version. A 0.7.0 file cannot be made compatible with Dify 1.15.x by ignoring the warning or changing only `version`: - upgrade the workspace to Dify 1.16.0 or later; or - ask the agent to generate a real DSL 0.6.0 workflow for Dify 1.15.x. The skill supports an explicit target version. For example: ```text Use $dify-workflow-dsl to create this workflow for Dify 1.15.0 using DSL 0.6.0. ``` Validate the result against the same target: ```bash python3 scripts/run_validator.py --strict --dify-version 1.15 workflow.yml ``` > [!TIP] > **New apps default to `workflow`.** Use or recommend `advanced-chat` only when > the requirement needs multi-turn memory, `sys.query`, chat file upload, or > `answer` nodes. ## ✨ What It Does - Generates `workflow`, `advanced-chat`, and DSL 0.7.0 Agent App YAML. - Supports portable Agent v2 packages, bindings, jobs, declared outputs, and omitted assets. - Covers Start, End, Answer, LLM, Code, IF/ELSE, Question Classifier, Human Input, Iteration, Loop, Assigner v2, tools, triggers, retrieval, files, and other common nodes. - Repairs graph wiring, branch handles, selectors, dependencies, and version-specific structures. - Reviews existing DSL with stable diagnostic codes and machine-readable JSON. - Keeps 0.6.0 compatibility for Dify 1.15.x. The skill treats tagged Dify source as schema authority. Exported files from the target workspace remain the authority for dynamic plugin/tool fields. ## 🧪 Dify 1.16 Evaluation Ten maintained DSL 0.7.0 scenarios cover the main static compatibility surface: | Scenario | Coverage | | --- | --- | | Text summarizer | Start → LLM → End | | Multi-turn assistant | Chatflow, `sys.query`, memory, Answer | | Excel analysis | File input, Document Extractor, Markdown conversion, LLM | | Priority routing | IF/ELSE and Variable Aggregator | | Question classification | Three classifier branches | | Array iteration | Iteration container and internal child topology | | Quality loop | Loop variables and Assigner v2 | | Human approval | Select/file-list inputs and action branches | | Agent v2 workflow | Package, binding, job, declared output | | Agent App | `app.mode: agent` and portable Soul | All ten pass: 1. the repository validator in strict mode; and 2. Dify 1.16.0's own production node registry, Human Input graph check, `AgentPackage`, and `WorkflowNodeJobConfig` models; and 3. Dify 1.16.0's real `AppDslService.import_app` against isolated PostgreSQL and Redis: **10 completed, 0 failed, 0 warnings**. This evaluation exposed and fixed a real issue: an older valid fixture omitted the LLM `context` object required by Dify. See the [full evaluation report](./references/dify-1.16-evaluation.md). ### Workflow Previews The two largest maintained graph examples each contain six nodes. Both files were imported through the real Dify 1.16.0 Console, then captured from Dify's Workflow canvas. Click an image to inspect the imported source workflow. #### Priority Routing [![Priority routing workflow with IF/ELSE branches and a Variable Aggregator](./assets/workflow-previews/04-priority-routing.png)](./examples/dify-1.16.0/04-priority-routing.yml) #### Array Iteration [![Array iteration workflow with an Iteration container and internal child nodes](./assets/workflow-previews/06-array-iteration.png)](./examples/dify-1.16.0/06-array-iteration.yml) ## 🤖 Agent v2 / DSL 0.7.0 The skill supports both 0.7.0 Agent forms: - top-level Agent App: `app.mode: agent` + `agent` + `agent_packages`; - Workflow/Chatflow Agent v2 node: `version: "2"` + `agent_binding.package_ref` + `agent_job`. Agent packages are portable, but secrets and workspace-bound assets are not. After import, review warnings and reconnect model credentials, tools, contacts, skills, files, and other omitted assets. The following screenshot is the checked-in Agent v2 example after a real import into Dify 1.16.0. The selected node and its inline Agent configuration panel are rendered by Dify itself. [![Agent v2 workflow node and inline configuration panel in Dify 1.16.0](./assets/workflow-previews/09-agent-v2-workflow.png)](./examples/dify-1.16.0/09-agent-v2-workflow.yml) ## 📦 Install ```bash git clone https://github.com/yzmw123/dify-workflow-dsl-skill.git cd dify-workflow-dsl-skill bash install.sh --platform codex --setup ``` Supported targets: ```bash bash install.sh --platform claude bash install.sh --platform codex bash install.sh --platform openclaw bash install.sh --platform hermes bash install.sh --platform opencode bash install.sh --platform all ``` Use `--target-dir` for a custom skills directory and `--force` to replace an existing installation. ## 🚀 Use Create a workflow: ```text Use $dify-workflow-dsl to create a Dify 1.18 workflow. Read an uploaded Excel file, convert the table to Markdown, analyze it with an LLM, and return the Markdown table plus the analysis. ``` Create an Agent v2 workflow: ```text Use $dify-workflow-dsl to create a DSL 0.7.0 workflow with a portable Agent v2 node, declared outputs, package refs, and strict validation. ``` Repair an existing file: ```text Use $dify-workflow-dsl to review and fix this YAML. Preserve its supported DSL version and report every import blocker and remaining workspace-specific step. ``` For an unfamiliar plugin tool, provide a minimal export from your workspace. That export is more reliable than a marketplace page or tool name alone. ## ✅ Validate Python 3.10+ is required. The runner installs PyYAML into a cached virtual environment on first use and reuses it thereafter. No global Python installation is modified. `install.sh --setup` prepares the runtime in advance. ```bash python3 scripts/run_validator.py --strict --dify-version 1.18 workflow.yml python3 scripts/run_validator.py --format json --strict --dify-version 1.18 workflow.yml python3 scripts/run_validator.py --strict --dify-version 1.16 old-workflow.yml python3 scripts/run_validator.py --strict --dify-version 1.15 old-0.6.yml ``` Set `DIFY_DSL_VENV_DIR` to select a runtime directory. First setup needs a Python package index; offline machines can install local wheels and set `DIFY_DSL_OFFLINE=1`. See [runtime setup](./references/runtime.md). The original `validate_dsl.py` entrypoint remains available in prepared Python environments. Run development regression tests in a virtual environment: ```bash python3 -m venv .venv .venv/bin/python -m pip install -r requirements-dev.txt .venv/bin/python -m unittest discover -s tests -v ``` Source validation uses the target Dify API environment (1.18: Python 3.13; 1.16: Python 3.12): ```bash /path/to/dify-1.18.0/api/.venv/bin/python scripts/validate_with_dify_source.py \ --dify-source /path/to/dify-1.18.0 --dify-version 1.18.0 \ examples/dify-1.16.0/*.yml examples/dify-1.18.0/*.yml ``` New checks cover Agent routes and their edges/selectors, reserved outputs, workspace skills, KnowledgeFS bindings and retrieval. Existing graph checks are retained. Import `.ifpkg` through Dify, or explicitly export YAML for local validation. Official difyctl handles online app operations; this skill authors and validates YAML. See [1.18 guidance](./references/dify-1.18-target.md). ## 🗂️ Repository Map ```text . ├── SKILL.md # Agent-facing operating rules ├── examples/dify-1.16.0/ # 10 maintained DSL 0.7.0 scenarios ├── examples/dify-1.18.0/ # 4 Dify 1.18 scenarios ├── references/ # Source-backed schemas and patterns ├── scripts/ │ ├── dify_dsl_validator/ # Reusable validator package │ ├── run_validator.py # Isolated dependency bootstrap │ ├── validate_dsl.py # Local deterministic CLI │ └── validate_with_dify_source.py # Dify-source compatibility gate ├── tests/ # Fixtures and regression tests ├── install.sh ├── README.md └── README_CN.md ``` ## ⚠️ Validation Boundary Static validation cannot prove that a workflow will run in every workspace. Real Dify import and execution are still required for: - installed plugin versions and dynamic tool schemas; - model availability and credentials; - knowledge bases, contacts, skills, files, and private assets; - external APIs, Human Input delivery, and side effects; - UI save/re-export behavior. No credentials belong in public DSL. Prefer fixed, parameterized SQL over model-generated mutation or DDL. ## 📚 Research Basis The project was built from Dify's tagged source, personal exports, and 262 parseable public App DSL files. Public examples are useful for graph patterns and backward compatibility, but older files are not treated as current schema authority. Primary links: - [Dify](https://github.com/langgenius/dify) - [Dify Marketplace](https://marketplace.dify.ai/) - [Dify official plugins](https://github.com/langgenius/dify-official-plugins) - [Agent Skills specification](https://agentskills.io/specification) ## Follow Silicon Scout S01 On WeChat Welcome to follow my WeChat Official Account: **硅基斥候 S01 (Silicon Scout S01)**. The account is still just getting started, so if this skill helps you, a follow would mean a lot. I will keep sharing practical skills I personally find useful, plus AI news, knowledge, and notes. **Recon focus** Silicon Scout S01 follows large language models, AI agents, AI coding and toolchains, hands-on product tests, enterprise and government AI adoption, and policy, security, and compliance. **Why "Scout"?** A scout enters unknown territory first, checks what is real, and brings back useful intelligence. In the AI context, S01 does that by personally testing new models, tools, and products, checking their risk boundaries, and explaining the actionable takeaways clearly. My other open-source projects will also be announced on this account first. <img src="./assets/wechat-official-account.jpg" alt="WeChat Official Account QR code for Silicon Scout S01" width="220">

저장소 README를 불러오는 중…

이 저장소의 skills

분류 대기 중

skill

직업 분류

설명

업데이트

dify-workflow-dsl

미분류

Create, modify, review, migrate, and debug import-ready Dify App DSL YAML for Dify 1.18 and legacy 1.15/1.16 targets (DSL 0.6.0 and 0.7.0). Use for Workflow, Chatflow, Agent App, Agent v2 workflow nodes, Human Input, graph wiring, variables, plugin…

원문 언어: 영어

2026-10-30

수집된 skill 1개 중 1개를 표시합니다.