feat: add OpenCode skills configuration and generation pipeline (fixes #74)

[hnshah]

Summary

Adds first-class OpenCode integration for this repository so OpenCode can discover and load all skills consistently with other agents.

Fixes #74

---

Changes

1. New Generator Script

scripts/generate_opencode_config.py

• Discovers all skills from skills/*/SKILL.md
• Extracts names, descriptions, and paths from YAML frontmatter
• Reuses .claude-plugin/plugin.json for metadata (version, description)
• Pulls MCP server URL from gemini-extension.json
• Generates opencode.json with complete configuration
• Supports --check flag for validation

Follows same pattern as generate_cursor_plugin.py:

• PEP 723 script format
• Self-contained, no dependencies
• Run with uv run scripts/generate_opencode_config.py

2. Generated Configuration

opencode.json includes:

• Skill discovery settings (skills/ directory, agent-skills format)
• All 13 skill entries with full metadata
• AGENTS context file reference (agents/AGENTS.md)
• MCP server configuration (aligned with Gemini/Cursor)

Structure:

{
  "name": "huggingface-skills",
  "version": "1.0.1",
  "skills": {
    "discovery": { "locations": ["skills/"], "format": "agent-skills" },
    "entries": [ /* all 13 skills */ ]
  },
  "context": { "agentsFile": "agents/AGENTS.md" },
  "mcpServers": { /* HF MCP server */ }
}

3. Updated Generation Pipeline

scripts/publish.sh

• Added opencode.json to GENERATED_FILES array
• Integrated generate_opencode_config.py into run_generate()
• Added --check validation for OpenCode artifacts
• Updated help text to list opencode.json

4. Documentation

README.md

• New "OpenCode" section after Cursor
• Explains auto-generated configuration
• Lists what’s included (discovery, entries, context, MCP)
• Documents regeneration with ./scripts/publish.sh
• Notes consistency with other agent formats

---

Design Goals (from issue #74)

✅ Reproducible: Auto-generated from single source of truth
✅ Consistent: Aligned with Claude/Cursor/Gemini integrations
✅ Validated: --check flag prevents drift
✅ Documented: Clear README section for users/contributors
✅ Maintainable: Follows existing generator patterns

---

Testing

Generation:

./scripts/publish.sh
# Generated: /tmp/hf-skills/opencode.json
# Publish artifacts generated successfully.

Validation:

./scripts/publish.sh --check
# All generated artifacts are up to date.

Output verification:

• ✅ JSON structure valid
• ✅ All 13 skills included with correct metadata
• ✅ MCP server URL matches gemini-extension.json
• ✅ AGENTS context file referenced correctly
• ✅ Skill paths resolve to actual directories

---

Benefits

For OpenCode users:

• Automatic skill discovery from this repository
• Consistent experience with other coding agents
• Access to all 13 Hugging Face skills

For contributors:

• No manual opencode.json maintenance
• Validation prevents accidental drift
• Same workflow as other agent configurations

For maintainers:

• Reduced onboarding friction
• Single source of truth (SKILL.md frontmatter)
• Easy to keep in sync with skill additions/changes

---

Implementation Notes

Generator follows established patterns:

• Reuses parse_frontmatter() pattern from generate_cursor_plugin.py
• Self-contained PEP 723 script (no external deps)
• Clean error handling with meaningful messages
• JSON output with pretty formatting

Integration is minimal:

• 3 lines added to publish.sh generation
• 2 lines added to validation check
• 1 new entry in help text
• No changes to existing skills or infrastructure

---

Checklist

• Added scripts/generate_opencode_config.py
• Generated opencode.json with all skills
• Updated scripts/publish.sh (generate + check + help)
• Documented OpenCode usage in README.md
• Tested generation and validation
• Verified JSON structure and content
• Followed existing code patterns
• No breaking changes to existing workflows

---

Ready for review! 🚀

[burtenshaw]

Thanks for this.