OpenClaw: A Practical Guide to Running a Local Autonomous AI Agent — Part 3: Skills
Important skills
-
Context7 - Up-to-date Docs for AI Agents. Context7 pulls up-to-date, version-specific documentation and code examples straight from the source — and gives them to your coding agent.
# Add Context7 as an MCP server in OpenClaw, then authorize it. openclaw mcp add context7 \ --url https://mcp.context7.com/mcp/oauth \ --transport streamable-http \ --auth oauth openclaw mcp login context7 # Then check that OpenClaw can reach the server and see its tools: openclaw mcp status openclaw mcp probe context7 openclaw mcp tools context7 Use the Context7 tool to look up the current official documentation for the Ollama Python library. Then, write a small Python script that sends a prompt to a locally running Ollama model using the ollama package and prints the response. Use the documented API, state any setup requirements, and tell me whether Context7 was actually called. If Context7 is unavailable, say so instead of pretending you used it. -
Agent Skills. Production-grade engineering skills for AI coding agents.
DEFINE PLAN BUILD VERIFY REVIEW SHIP ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ │ Idea │ ───▶│ Spec │ ───▶│ Code │ ───▶│ Test │ ───▶│ QA │ ───▶│ Go │ │Refine│ │ PRD │ │ Impl │ │Debug │ │ Gate │ │ Live │ └──────┘ └──────┘ └──────┘ └──────┘ └──────┘ └──────┘ /spec /plan /build /test /review /ship -
Superpowers is a complete software development methodology for your coding agents, built on top of a set of composable skills and some initial instructions that make sure your agent uses them.
Using Images with OpenClaw and OpenRouter Free Models
The image_generate tool in OpenClaw is provider-agnostic: it routes requests to whichever image backend you have configured.
Generating images:
/tool image_generate action=generate prompt="A cinematic, reverent depiction of Jonah on the shore after being delivered by the great fish" aspectRatio=16:9 count=1
Verify that the tool is visible
To check which image providers and models are configured, use:
/tool image_generate action=list
Available image generation providers:
openrouter — configured ✅ (default: google/gemini-3.1-flash-image-preview)
[...]
Use the Async Pipeline Correctly
OpenClaw’s image_generate is asynchronous in chat sessions. The tool records a background task, returns a task ID immediately, and wakes the agent when the provider finishes. This is not an error state; it is the designed behavior.
---
name: daily-gospel
description: Fetch the daily Gospel reading and present it as a short guided prayer.
---
When the user asks for the daily Gospel or daily reading, run:
`{baseDir}/.venv/bin/python3 {baseDir}/scripts/daily_gospel.py`
The script prints a JSON object with `title`, `date`, `reading`, and `link`.
If it reports an error, exits non-zero, or returns no entry, say briefly that
the reading is unavailable today and stop — do not invent or substitute
scripture.
[...]
## Structure
1. **Opening — 🙏 (verbatim)**
Begin with this exact text, not a paraphrase:
> 🙏 Thank You for bringing us together to pray.
> [... It has already explain previously]
5. **Image — 🖼️**
Craft a descriptive, cinematic, and reverent prompt based on the core scene, imagery, or theme of the `reading`.
Invoke the image generator by outputting this exact tool command on its own line:
`/tool image_generate action=generate prompt="<cinematic prompt>" aspectRatio=16:9 count=1 && /tool image_generate action=download && /tool image`
**CRITICAL ASYNC RULE:** The image tool runs in the background. It will not return the image immediately. DO NOT apologize or say that you cannot generate images. Simply output the tool command, print a brief placeholder like `🖼️ *Visualizing today's reading...*`, and proceed immediately to the next step.
[... Continue]
## Reference format
[... Continue]
/tool image_generate action=generate prompt="<cinematic prompt>" aspectRatio=16:9 count=1 && /tool image_generate action=download && /tool image
🖼️ *Visualizing today's reading...*
[... Continue]
To generate, download, and display an image in one flow, output the tool command on its own line and then proceed immediately: /tool image_generate action=generate prompt="<cinematic prompt>" aspectRatio=16:9 count=1 && /tool image_generate action=download && /tool image