Add OpenCode skills config and generation pipeline

.cursor-plugin/plugin.json

@@ -3,7 +3,7 @@
   "skills": "skills",
   "mcpServers": ".mcp.json",
   "description": "Agent Skills for AI/ML tasks including dataset creation, model training, evaluation, and research paper publishing on Hugging Face Hub",
-  "version": "1.0.1",
+  "version": "1.0.0",
   "author": {
     "name": "Hugging Face"
   },

README.md

@@ -16,7 +16,7 @@ In practice, skills are self-contained folders that package instructions, script
 
 ## Installation
 
-Hugging Face skills are compatible with Claude Code, Codex, Gemini CLI, and Cursor.
+Hugging Face skills are compatible with Claude Code, Codex, Gemini CLI, Cursor, and OpenCode.
 
 ### Claude Code
 
@@ -46,6 +46,18 @@ For example:
 
 3. If your Codex setup still relies on `AGENTS.md`, you can use the generated [`agents/AGENTS.md`](agents/AGENTS.md) file in this repo as a fallback bundle of instructions.
 
+### OpenCode
+
+1. This repo includes an OpenCode config file at [`opencode.json`](opencode.json), generated from the same metadata sources as other agent integrations.
+
+2. Install or reference this repository in OpenCode using your normal extension/plugin workflow, then point OpenCode to the repo root so it can read:
+   - `opencode.json`
+   - `skills/`
+   - `agents/AGENTS.md`
+   - `.mcp.json`
+
+3. For field-level details and validation workflow, see [`docs/opencode.md`](docs/opencode.md).
+
 ### Gemini CLI
 
 1. This repo includes `gemini-extension.json` to integrate with the Gemini CLI.
@@ -79,6 +91,8 @@ For contributors, regenerate manifests with:
 ./scripts/publish.sh
 ```
 
+This also regenerates `opencode.json`.
+
 ## Skills
 
 This repository contains a few skills to get you started. You can also contribute your own skills to the repository.

agents/AGENTS.md

@@ -32,4 +32,6 @@ hugging-face-trackio: `Track and visualize ML training experiments with Trackio.
 
 Paths referenced within SKILL folders are relative to that SKILL. For example the hf-datasets `scripts/example.py` would be referenced as `hf-datasets/scripts/example.py`. 
 
+OpenCode compatibility: this repository also ships `opencode.json` at repo root. It points OpenCode to `skills/`, `agents/AGENTS.md`, and MCP servers so OpenCode can load all skills consistently with other agents.
+
 </skills>

docs/opencode.md

@@ -0,0 +1,47 @@
+# OpenCode Integration
+
+This repository provides an OpenCode-compatible config at `opencode.json`.
+
+## What `opencode.json` contains
+
+- `name`, `description`, `author`, `homepage`, `repository`, `license`, `keywords`
+  - Sourced from `.claude-plugin/plugin.json` to keep metadata aligned.
+- `version`
+  - OpenCode config version (currently independent, set to `0.1.0`).
+- `skills`
+  - Set to `skills` so OpenCode can scan all skill folders.
+- `skillPaths`
+  - Enumerates all discovered skill directories from `skills/*/SKILL.md`.
+- `contextFileName`
+  - Defaults to `agents/AGENTS.md`, aligned with `gemini-extension.json` when present.
+- `mcpServers`
+  - Pulled from `.mcp.json` (fallback to `gemini-extension.json` MCP data when needed).
+
+## Generation and validation
+
+`opencode.json` is generated, not hand-edited.
+
+Use:
+
+```bash
+uv run scripts/generate_opencode_config.py
+```
+
+Validate without writing:
+
+```bash
+uv run scripts/generate_opencode_config.py --check
+```
+
+Or run the full publish pipeline:
+
+```bash
+./scripts/publish.sh
+./scripts/publish.sh --check
+```
+
+## Why this design
+
+- Reuses existing single sources of truth for metadata and MCP settings.
+- Keeps OpenCode integration in the same generation/check lifecycle as AGENTS/README/Cursor artifacts.
+- Automatically tracks all skills under `skills/` with no manual list maintenance.

opencode.json

@@ -0,0 +1,40 @@
+{
+  "name": "huggingface-skills",
+  "description": "Agent Skills for AI/ML tasks including dataset creation, model training, evaluation, and research paper publishing on Hugging Face Hub",
+  "version": "0.1.0",
+  "skills": "skills",
+  "skillPaths": [
+    "skills/hf-cli",
+    "skills/hugging-face-dataset-viewer",
+    "skills/hugging-face-datasets",
+    "skills/hugging-face-evaluation",
+    "skills/hugging-face-jobs",
+    "skills/hugging-face-model-trainer",
+    "skills/hugging-face-paper-publisher",
+    "skills/hugging-face-tool-builder",
+    "skills/hugging-face-trackio",
+    "skills/huggingface-gradio"
+  ],
+  "contextFileName": "agents/AGENTS.md",
+  "mcpServers": {
+    "huggingface-skills": {
+      "url": "https://huggingface.co/mcp?login"
+    }
+  },
+  "homepage": "https://github.com/huggingface/skills",
+  "repository": "https://github.com/huggingface/skills",
+  "license": "Apache-2.0",
+  "keywords": [
+    "huggingface",
+    "machine-learning",
+    "datasets",
+    "training",
+    "evaluation",
+    "papers",
+    "fine-tuning",
+    "llm"
+  ],
+  "author": {
+    "name": "Hugging Face"
+  }
+}

scripts/AGENTS_TEMPLATE.md

@@ -19,4 +19,6 @@ IMPORTANT: You MUST read the SKILL.md file whenever the description of the skill
 
 Paths referenced within SKILL folders are relative to that SKILL. For example the hf-datasets `scripts/example.py` would be referenced as `hf-datasets/scripts/example.py`. 
 
+OpenCode compatibility: this repository also ships `opencode.json` at repo root. It points OpenCode to `skills/`, `agents/AGENTS.md`, and MCP servers so OpenCode can load all skills consistently with other agents.
+
 </skills>

scripts/generate_opencode_config.py

@@ -0,0 +1,150 @@
+#!/usr/bin/env -S uv run
+# /// script
+# requires-python = ">=3.10"
+# dependencies = []
+# ///
+"""Generate OpenCode config from existing repo metadata.
+
+Output:
+- opencode.json
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+from pathlib import Path
+
+
+ROOT = Path(__file__).resolve().parent.parent
+CLAUDE_PLUGIN_MANIFEST = ROOT / ".claude-plugin" / "plugin.json"
+GEMINI_EXTENSION = ROOT / "gemini-extension.json"
+MCP_CONFIG = ROOT / ".mcp.json"
+OPENCODE_CONFIG = ROOT / "opencode.json"
+
+OPENCODE_CONFIG_VERSION = "0.1.0"
+DEFAULT_CONTEXT_FILE = "agents/AGENTS.md"
+
+
+def load_json(path: Path) -> dict:
+    if not path.exists():
+        raise FileNotFoundError(f"Missing required file: {path}")
+    return json.loads(path.read_text(encoding="utf-8"))
+
+
+def collect_skill_paths() -> list[str]:
+    paths: list[str] = []
+    for skill_md in sorted(ROOT.glob("skills/*/SKILL.md")):
+        paths.append(str(skill_md.parent.relative_to(ROOT)))
+    return paths
+
+
+def extract_context_file() -> str:
+    if not GEMINI_EXTENSION.exists():
+        return DEFAULT_CONTEXT_FILE
+
+    data = load_json(GEMINI_EXTENSION)
+    context_file = data.get("contextFileName")
+    if not isinstance(context_file, str) or not context_file.strip():
+        return DEFAULT_CONTEXT_FILE
+
+    return context_file
+
+
+def extract_mcp_servers() -> dict:
+    if MCP_CONFIG.exists():
+        data = load_json(MCP_CONFIG)
+        mcp_servers = data.get("mcpServers")
+        if isinstance(mcp_servers, dict) and mcp_servers:
+            return mcp_servers
+
+    if GEMINI_EXTENSION.exists():
+        data = load_json(GEMINI_EXTENSION)
+        mcp_servers = data.get("mcpServers")
+        if isinstance(mcp_servers, dict) and mcp_servers:
+            normalized = {}
+            for server_name, cfg in mcp_servers.items():
+                if not isinstance(cfg, dict):
+                    continue
+                url = cfg.get("url") or cfg.get("httpUrl")
+                if isinstance(url, str) and url.strip():
+                    normalized[server_name] = {"url": url}
+            if normalized:
+                return normalized
+
+    return {}
+
+
+def build_opencode_config() -> dict:
+    plugin_manifest = load_json(CLAUDE_PLUGIN_MANIFEST)
+    skill_paths = collect_skill_paths()
+    if not skill_paths:
+        raise ValueError("No skills discovered under skills/*/SKILL.md")
+
+    name = plugin_manifest.get("name")
+    description = plugin_manifest.get("description")
+    if not isinstance(name, str) or not name:
+        raise ValueError(".claude-plugin/plugin.json must define a non-empty 'name'")
+    if not isinstance(description, str) or not description:
+        raise ValueError(".claude-plugin/plugin.json must define a non-empty 'description'")
+
+    config = {
+        "name": name,
+        "description": description,
+        "version": OPENCODE_CONFIG_VERSION,
+        "skills": "skills",
+        "skillPaths": skill_paths,
+        "contextFileName": extract_context_file(),
+        "mcpServers": extract_mcp_servers(),
+    }
+
+    for key in ["homepage", "repository", "license", "keywords", "author"]:
+        if key in plugin_manifest:
+            config[key] = plugin_manifest[key]
+
+    return config
+
+
+def render_json(data: dict) -> str:
+    return json.dumps(data, indent=2, ensure_ascii=False) + "\n"
+
+
+def write_or_check(path: Path, content: str, check: bool) -> bool:
+    current = path.read_text(encoding="utf-8") if path.exists() else None
+    if current == content:
+        return True
+
+    if check:
+        return False
+
+    path.write_text(content, encoding="utf-8")
+    return True
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Generate OpenCode config")
+    parser.add_argument(
+        "--check",
+        action="store_true",
+        help="Validate opencode.json is up-to-date without writing changes.",
+    )
+    args = parser.parse_args()
+
+    content = render_json(build_opencode_config())
+    ok = write_or_check(OPENCODE_CONFIG, content, check=args.check)
+
+    if args.check:
+        if not ok:
+            print("OpenCode config is out of date:", file=sys.stderr)
+            print(f"  - {OPENCODE_CONFIG.relative_to(ROOT)}", file=sys.stderr)
+            print("Run: uv run scripts/generate_opencode_config.py", file=sys.stderr)
+            sys.exit(1)
+        print("OpenCode config is up to date.")
+        return
+
+    print(f"Wrote {OPENCODE_CONFIG.relative_to(ROOT)}")
+
+
+if __name__ == "__main__":
+    main()

scripts/publish.sh

@@ -9,6 +9,7 @@ GENERATED_FILES=(
   "README.md"
   ".cursor-plugin/plugin.json"
   ".mcp.json"
+  "opencode.json"
 )
 
 file_sig() {
@@ -23,6 +24,7 @@ file_sig() {
 run_generate() {
   uv run scripts/generate_agents.py
   uv run scripts/generate_cursor_plugin.py
+  uv run scripts/generate_opencode_config.py
 }
 
 run_check() {
@@ -54,6 +56,7 @@ run_check() {
 
   # Extra explicit check for cursor-only artifacts
   uv run scripts/generate_cursor_plugin.py --check
+  uv run scripts/generate_opencode_config.py --check
 
   echo "All generated artifacts are up to date."
 }
@@ -77,6 +80,7 @@ This script regenerates:
   - README.md (skills table section)
   - .cursor-plugin/plugin.json
   - .mcp.json
+  - opencode.json
 EOF
     ;;
   *)