rlm-controller

What This Skill Does

The rlm-controller is a sophisticated orchestration layer designed for OpenClaw to handle information that exceeds standard model context windows. Instead of attempting to cram massive datasets or entire code repositories into a single prompt, this skill treats inputs as an external context store. It utilizes a modular system of helper scripts (e.g., rlm_ctx.py, rlm_auto.py) to store, index, slice, and intelligently query vast amounts of data. By spawning localized subcalls through OpenClaw’s sessions_spawn tool, the controller can perform high-density analysis while adhering to strict recursion and safety boundaries. It is designed to act as a "manager" that delegates specific analytical tasks to sub-agent sessions, aggregating findings back into a coherent, structured response for the user.

Installation

To integrate this skill into your environment, use the OpenClaw hub command: clawhub install openclaw/skills/skills/skywyze/rlm-controller Ensure your agent environment has access to the standard set of OpenClaw system tools (read, write, exec) as the controller relies on these to interface with its bundled safety-checked helper scripts.

Use Cases

• Repository-scale Code Audits: Analyze deep dependency trees or security patterns across thousands of lines of code.
• Large-scale Log Analysis: Ingest gigabyte-sized server logs to find anomalies or trace specific transaction IDs.
• Document Synthesis: Process multiple PDF or markdown manuals simultaneously to extract answers that require cross-referencing information distributed across files.
• Dataset Exploration: Perform statistical summaries or qualitative analysis on structured datasets that are too large for a single-pass inference.

Example Prompts

1. "Use the rlm-controller to analyze the /data/logs/prod-v2/ directory and identify any repeated 500-error patterns from the last 24 hours."
2. "I need to map out the authentication flow in this repo; use rlm-controller to scan all files in the /src/auth folder and generate a sequence diagram summary."
3. "Analyze these 50 technical manuals stored in /docs/manuals and answer the following question: 'What is the required configuration for the secondary failover node?'"

Tips & Limitations

• Safety First: Always consult docs/policy.md before adjusting default limits. The controller enforces a max recursion depth of 1, preventing runaway agent chains.
• Cleanup: The tool creates temporary artifacts during the indexing process. Regularly run scripts/cleanup.sh to prevent disk bloat, especially when processing large datasets.
• Non-blocking Spawns: Remember that sessions_spawn is asynchronous. The controller is designed to manage these results, so avoid manually killing sub-sessions while a run is in progress.
• Configuration: If you require human-in-the-loop validation for every sub-task, set disableModelInvocation: true in your OpenClaw config to ensure you approve every sessions_spawn call.