Lazy load prompts on demand for OpenCode agents. Operations teams benefit by dynamically injecting skills during conversations. Connects to Claude agents, enhancing workflow automation.
git clone https://github.com/zenobi-us/opencode-skillful.gitOpenCode Skills Plugin implements the Anthropic Agent Skills Specification with lazy-loaded skill discovery and injection. Unlike built-in OpenCode that pre-loads all skills, this plugin loads skills only when explicitly requested, reducing token consumption and context bloat. Agents use three core tools—skill_find to discover skills by keyword, skill_use to load skills into chat, and skill_resource to access specific documentation or templates—enabling dynamic skill management during conversations. This approach is particularly effective for operations teams managing 50+ skills where only a few are needed per conversation. The plugin also supports per-model format configuration, allowing you to optimize skill formatting for different LLM providers without creating duplicate documentation.
["Identify the skill you want to dynamically load (e.g., `data-processing`, `report-generation`). Ensure the skill is compatible with the opencode-skillful system.","In your OpenCode agent configuration, enable the opencode-skillful feature and specify the skill repository or directory where skills are stored. Example for Claude: `Enable opencode-skillful and set skill_repository to /path/to/skills`.","Use the prompt template to trigger on-demand skill loading. Replace `[SKILL_NAME]` with the skill you need (e.g., `report-generation`) and `[PARAMETERS]` with the required inputs (e.g., `template: monthly_report.md`, `data_source: sales_data.csv`).","Monitor the agent's response for confirmation of skill loading and execution. Review the structured output for results and next steps.","For advanced use, chain multiple skills by triggering subsequent prompts after the first skill completes. Example: After `data-processing`, trigger `report-generation` with the cleaned dataset as input."]
Search and load git commit best practices during code reviews
Dynamically inject testing templates without pre-loading all skills
Access specific skill documentation or guides mid-conversation
Manage large skill libraries with minimal token overhead
No install command available. Check the GitHub repository for manual installation instructions.
git clone https://github.com/zenobi-us/opencode-skillfulCopy the install command above and run it in your terminal.
Launch Claude Code, Cursor, or your preferred AI coding agent.
Use the prompt template or examples below to test the skill.
Adapt the skill to your specific use case and workflow.
Initialize the OpenCode agent with the following prompt template for dynamic skill injection: 'Act as an OpenCode agent with the ability to dynamically load and execute the [SKILL_NAME] skill when triggered by the user. The skill should be loaded on-demand using the opencode-skillful system. Once loaded, execute the skill with the following parameters: [PARAMETERS]. Respond with the results in a structured format.' Ensure the agent acknowledges the dynamic loading and confirms the skill execution.
```json
{
"status": "success",
"skill_loaded": "data-processing",
"execution_result": {
"processed_records": 1247,
"errors": 3,
"avg_processing_time_ms": 45,
"sample_output": [
{"id": "REC-001", "status": "cleaned", "changes": ["removed_special_chars", "normalized_case"]},
{"id": "REC-002", "status": "skipped", "reason": "duplicate_record"},
{"id": "REC-003", "status": "cleaned", "changes": ["trimmed_whitespace"]}
]
},
"next_steps": [
"Review the 3 errors for manual correction",
"Log the processing metrics to the data pipeline dashboard",
"Trigger downstream validation if all records are processed successfully"
]
}
```
The OpenCode agent dynamically loaded the `data-processing` skill using the opencode-skillful system. Upon receiving the user's request to clean a dataset of 1,250 records, the agent first verified the skill's availability and then executed it with the provided parameters (e.g., `input_dataset: sales_leads_q3.csv`, `validation_rules: standard_cleaning`). The skill processed the records in batches, logging errors for duplicates and special characters. The agent then formatted the results into a structured JSON response, including key metrics and actionable next steps for the operations team. This dynamic loading reduced the need for pre-loading all skills, optimizing memory usage and enabling faster response times during peak workloads.AI sales agent for lead generation and follow-up
AI-assisted web application security testing
Get more done every day with Microsoft Teams – powered by AI
Automate your spreadsheet tasks with AI power
Agentic AI Workflow platform
Connected workspace for docs, wikis, and projects
Take a free 3-minute scan and get personalized AI skill recommendations.
Take free scan