agentql-automation
by ComposioHQagentql-automation helps Claude run AgentQL browser automation through Composio Rube MCP with schema-first tool discovery, connection checks, and safer execution steps.
This skill scores 68/100, which means it is acceptable for listing but should be presented as a lightweight integration guide rather than a fully packaged workflow skill. Directory users get enough information to understand that it routes AgentQL automation through Rube MCP and how an agent should discover tools and verify connections, but the repository evidence shows limited concrete task examples and no support files, so adoption may require extra experimentation.
- Valid skill frontmatter declares the required `rube` MCP and clearly frames when to use it: automating AgentQL operations through Composio/Rube.
- Prerequisites and setup steps are explicit, including checking `RUBE_SEARCH_TOOLS`, using `RUBE_MANAGE_CONNECTIONS`, and confirming an ACTIVE AgentQL connection.
- The skill instructs agents to search tools first for current schemas, which improves triggerability and reduces stale-tool guesswork.
- No support files or install command are provided; setup depends on manually adding the Rube MCP endpoint and configuring an AgentQL connection.
- The workflow guidance is mostly meta-level tool discovery and connection checking, so users may need additional AgentQL/domain knowledge for specific automation tasks.
Overview of agentql-automation skill
What agentql-automation is for
agentql-automation is a Claude skill for running AgentQL browser automation tasks through Composio’s Rube MCP server. It is designed for workflows where an AI agent needs to discover the current AgentQL tool schema, verify an active AgentQL connection, and then execute browser-oriented operations using the Rube tool layer instead of guessing tool names or stale parameters.
The main job-to-be-done is not “write a browser automation script from scratch.” It is to help an agent call the right Composio AgentQL tools in the right order, with connection checks and schema discovery before execution.
Best-fit users and projects
This skill fits users who already use Claude with MCP tools and want AgentQL access through Rube. It is most useful for automation tasks involving web pages, structured extraction, browser actions, or AgentQL-powered browsing where tool availability may change over time.
It is a good install candidate if you want a repeatable prompting pattern for AgentQL via Composio. It is less useful if you do not use Rube MCP, do not have an AgentQL connection, or only need static code examples without live tool execution.
Key differentiator: schema-first execution
The important behavior in the agentql-automation skill is its insistence on calling RUBE_SEARCH_TOOLS first. That matters because Rube can return current tool slugs, input schemas, recommended execution plans, and known pitfalls. For browser automation, this reduces failures caused by outdated assumptions about tool names or arguments.
The skill also makes connection status part of the workflow by using RUBE_MANAGE_CONNECTIONS with toolkit agentql before running AgentQL operations.
How to Use agentql-automation skill
agentql-automation install and setup context
Install the skill from the Composio skills repository, then make sure your client can use Rube MCP:
npx skills add ComposioHQ/awesome-claude-skills --skill agentql-automation
The skill expects Rube MCP to be configured as an MCP server using:
https://rube.app/mcp
Before relying on the skill, confirm that RUBE_SEARCH_TOOLS is available. Then use RUBE_MANAGE_CONNECTIONS with toolkit agentql. If AgentQL is not active, follow the returned authentication link and rerun the connection check until the status is ACTIVE.
Inputs the skill needs for good execution
A weak prompt is: “Use AgentQL to automate this site.”
A stronger prompt gives the task, target, success criteria, and constraints:
“Use agentql-automation for Browser Automation. First search Rube tools for the current AgentQL schema. Then check the AgentQL connection. On https://example.com/products, find all visible product names and prices from the first results page. Return structured JSON with name, price, and source_url. Do not purchase, submit forms, or bypass login.”
This improves output because the agent can convert the goal into a specific RUBE_SEARCH_TOOLS query, choose the right AgentQL tool, and avoid unsafe or ambiguous browser actions.
Recommended workflow
Use this sequence when invoking the agentql-automation skill:
- Ask the agent to call
RUBE_SEARCH_TOOLSfor the exact use case, not a generic “AgentQL operations” query. - Reuse the returned session ID where appropriate so the execution plan stays coherent.
- Call
RUBE_MANAGE_CONNECTIONSforagentqlbefore attempting any browser operation. - If active, run the selected AgentQL tool using the discovered schema.
- Ask the agent to report which tool slug, required fields, and assumptions it used.
This workflow is especially helpful when AgentQL tasks change from extraction to navigation, because each task may require different fields or execution plans.
Files to read before adoption
The repository path is composio-skills/agentql-automation, and the main file to inspect is SKILL.md. There are no visible helper scripts, resources, rules, or README files in the provided tree, so the install decision should be based on the skill instructions themselves.
Read SKILL.md for prerequisites, setup, tool discovery, and the core workflow pattern. Also check the official toolkit documentation linked from the skill: https://composio.dev/toolkits/agentql.
agentql-automation skill FAQ
Is agentql-automation enough without Rube MCP?
No. The skill explicitly requires the rube MCP server and depends on tools such as RUBE_SEARCH_TOOLS and RUBE_MANAGE_CONNECTIONS. Without those tools, it becomes only a workflow note and cannot execute AgentQL automation.
How is this better than an ordinary AgentQL prompt?
A generic prompt may ask the model to infer tool names or call AgentQL directly. The agentql-automation skill adds a safer operational sequence: discover available tools first, inspect the current schema, verify the AgentQL connection, then execute. That is valuable when schemas or recommended execution plans may change.
Is this suitable for beginners?
It is beginner-friendly only if the user is already comfortable with MCP-enabled clients. The browser automation goal can be simple, but setup still requires adding the Rube MCP endpoint, managing the AgentQL connection, and understanding that tool discovery is mandatory before execution.
When should I not use this skill?
Do not use it for non-AgentQL browser automation, offline scraping scripts, or tasks where no MCP tool execution is available. Also avoid it for workflows that require bypassing access controls, submitting sensitive forms without review, or operating on pages where you cannot define clear success criteria.
How to Improve agentql-automation skill
Make prompts more specific than the page goal
To get better results from agentql-automation, describe the browser task in terms of observable page actions and expected output. Include the URL, what to extract or change, required fields, stopping condition, and forbidden actions.
Better input:
“Open the target page, identify the pricing table, extract plan names, monthly prices, annual prices if visible, and feature limits. Return a markdown table. Stop after the first pricing page. Do not click checkout or login links.”
This gives the agent enough context to form a precise Rube tool search and validate completion.
Prevent common failure modes
The most common failure is skipping tool discovery and using assumed schemas. Make “call RUBE_SEARCH_TOOLS first” part of every request. Another common issue is running before AgentQL is connected; require the agent to confirm ACTIVE status before execution.
For dynamic pages, ask the agent to report uncertainty: missing elements, login walls, pagination, blocked content, or differences between expected and observed page structure.
Iterate after the first run
After the first output, refine based on what the tool returned. If fields are missing, ask for a narrower AgentQL query or a second pass on a specific page region. If the output is too broad, add a limit such as “first 10 results,” “visible cards only,” or “current page only.”
A strong iteration prompt is:
“Using the same AgentQL/Rube session if possible, rerun only the extraction step. Keep name and price, add availability, and ignore sponsored results.”
Strengthen the skill itself for teams
Teams adopting the agentql-automation skill can improve reliability by adding local examples for their common sites, approved action boundaries, expected output schemas, and connection troubleshooting notes. Because the upstream skill is compact and has no support files, team-specific guidance can add real value without changing the core schema-first workflow.