sourcegraph-automation
by ComposioHQsourcegraph-automation helps Claude agents automate Sourcegraph workflows through Composio Rube MCP by discovering current tool schemas, checking the Sourcegraph connection, and executing with verified inputs.
This skill scores 66/100, which makes it acceptable but limited for directory listing. Directory users can understand that it is for Sourcegraph automation through Rube MCP and get enough setup guidance to try it, but they should expect a lightweight wrapper around dynamic tool discovery rather than a deeply documented Sourcegraph workflow pack.
- Valid frontmatter and a clear description identify the trigger: automating Sourcegraph tasks through Rube MCP/Composio.
- Prerequisites and setup steps are explicit, including adding the Rube MCP endpoint, checking RUBE_SEARCH_TOOLS, and activating a Sourcegraph connection with RUBE_MANAGE_CONNECTIONS.
- The skill emphasizes tool discovery before execution, which should help agents avoid stale Sourcegraph schemas.
- No support files, scripts, references, or README are present beyond SKILL.md, so adoption depends entirely on the brief in-skill instructions.
- The workflow relies on dynamic RUBE_SEARCH_TOOLS discovery and does not appear to provide concrete Sourcegraph task examples or fixed tool schemas, leaving some execution guesswork.
Overview of sourcegraph-automation skill
What sourcegraph-automation does
sourcegraph-automation is a Claude skill for automating Sourcegraph tasks through Composio’s Rube MCP server. Instead of assuming fixed tool names or stale arguments, the skill directs the agent to search Rube’s current Sourcegraph tool schemas first, verify the Sourcegraph connection, and then execute the requested workflow with the correct tool inputs.
Use it when you want an AI agent to work with Sourcegraph through MCP for repository search, code intelligence, or Sourcegraph-related operations without manually looking up every Composio tool schema.
Best fit for Sourcegraph and MCP users
This sourcegraph-automation skill is best for teams already using Sourcegraph and willing to connect it through Rube MCP. It is especially useful for workflow automation where the exact Sourcegraph action may vary by environment, account permissions, or current Composio toolkit schema.
It fits users who need an agent to:
- discover available Sourcegraph tools before acting;
- check whether authentication is active;
- convert a natural-language Sourcegraph task into MCP tool calls;
- avoid brittle prompts that reference outdated tool names.
What makes it different from a generic prompt
A generic prompt might say “use Sourcegraph to search the codebase.” This skill adds a safer execution pattern: discover tools with RUBE_SEARCH_TOOLS, confirm the Sourcegraph connection with RUBE_MANAGE_CONNECTIONS, then call the tool that matches the current schema.
That distinction matters because MCP tool definitions can change. The skill’s most important rule is practical: do not guess tool parameters. Search first, then execute.
Important adoption constraints
The repository contains a single SKILL.md and no helper scripts, examples folder, or local test harness. Adoption depends on your MCP client supporting Rube and on your Sourcegraph connection being active through Composio.
If you are looking for a standalone CLI, a Sourcegraph API wrapper, or a batch migration script, this is not that. sourcegraph-automation is an agent instruction layer for MCP-driven Sourcegraph operations.
How to Use sourcegraph-automation skill
sourcegraph-automation install context
Install the skill in a Claude skills-compatible environment, for example:
npx skills add ComposioHQ/awesome-claude-skills --skill sourcegraph-automation
Then configure Rube MCP in your client by adding the MCP server endpoint:
https://rube.app/mcp
Before expecting the skill to work, confirm that the MCP tool RUBE_SEARCH_TOOLS is available. Next, use RUBE_MANAGE_CONNECTIONS with toolkit sourcegraph and complete the returned auth flow if the connection is not ACTIVE.
Inputs the skill needs from you
For reliable sourcegraph-automation usage, give the agent the actual job, not just the product name. Include:
- the Sourcegraph task, such as code search, repository lookup, or investigation;
- repository names, org names, code symbols, file paths, or search terms;
- the desired output format, such as summary, links, table, or action plan;
- safety limits, such as read-only mode or “do not modify settings”;
- whether the agent should stop after tool discovery or proceed to execution.
Weak prompt: “Use Sourcegraph to check auth code.”
Stronger prompt: “Using sourcegraph-automation, verify the Sourcegraph connection first, discover current tools, then search for OAuth token validation logic in acme/backend and summarize relevant files with Sourcegraph links. Do not make changes.”
Recommended workflow for first use
Start by reading composio-skills/sourcegraph-automation/SKILL.md; it is the only source file and contains the operating pattern. The key sequence is:
- Call
RUBE_SEARCH_TOOLSfor the specific Sourcegraph use case. - Reuse the returned session ID when continuing discovery or execution.
- Check the Sourcegraph connection through
RUBE_MANAGE_CONNECTIONS. - If inactive, complete the auth link and retry.
- Use only tool slugs and fields returned by discovery.
- Report the result and any tool limitations back to the user.
This is a good sourcegraph-automation guide pattern for first adoption because it separates setup failure from task failure.
Practical prompt pattern
Use prompts that force the skill to avoid schema guessing:
“Use the sourcegraph-automation skill for Workflow Automation. First call RUBE_SEARCH_TOOLS for: [specific Sourcegraph task]. If the Sourcegraph connection is inactive, stop and tell me what auth step is needed. If active, run the matching tool with the discovered schema. Return [format] and include any assumptions.”
This phrasing improves output quality because the agent knows when to stop, when to authenticate, and how to shape the final answer.
sourcegraph-automation skill FAQ
Is sourcegraph-automation only for Composio users?
Yes, in practice. The skill is built around Composio’s Rube MCP server and the Sourcegraph toolkit exposed through it. If your environment does not support MCP tools or cannot connect to Rube, the instructions will not have the required execution layer.
Can beginners use this skill?
Beginners can use it if their MCP client is already configured. The hard part is not the Sourcegraph concept; it is connection readiness. New users should first verify RUBE_SEARCH_TOOLS responds and that RUBE_MANAGE_CONNECTIONS shows an active Sourcegraph connection before asking for real work.
How is this different from using Sourcegraph directly?
Sourcegraph directly gives you UI and API access. sourcegraph-automation gives an AI agent a repeatable process for discovering available Sourcegraph MCP tools and using them correctly. It is most valuable when you want the agent to decide which current tool schema to use for a workflow.
When should I not install sourcegraph-automation?
Do not install it if you need offline operation, a non-MCP integration, a fully documented command-line utility, or extensive bundled examples. Also skip it if your organization cannot authorize Sourcegraph through Composio/Rube.
How to Improve sourcegraph-automation skill
Improve sourcegraph-automation results with task-specific context
The fastest way to improve sourcegraph-automation output is to provide searchable context. Include exact repository names, branches if relevant, symbols, error messages, file extensions, or Sourcegraph query constraints.
Better inputs reduce broad searches and prevent the agent from wasting tool calls on discovery that is technically correct but too general.
Avoid common failure modes
The most common failure is schema guessing. If the agent tries to call a Sourcegraph tool before RUBE_SEARCH_TOOLS, redirect it: “Search current Sourcegraph tool schemas first, then continue.”
Other common issues include inactive connections, vague repository targets, and prompts that ask for write-like actions without clarifying permissions. For safer use, say whether the task is read-only and whether the agent may proceed after authentication is confirmed.
Iterate after the first output
After the first run, ask for refinement based on concrete gaps:
- “Narrow this to repositories under
platform/only.” - “Return only files changed in the last release area.”
- “Convert the findings into a triage checklist.”
- “Run another discovery query for code navigation tools, then continue.”
This keeps the skill anchored to current tool availability while improving the final answer.
What would make the skill stronger
The repository would be more adoption-ready with a minimal example library: common Sourcegraph tasks, expected discovery prompts, sample connection-check flows, and failure-handling examples. Until then, users should treat SKILL.md as the source of truth and write prompts that explicitly require discovery, connection validation, and schema-faithful execution.