stream-of-consciousness

What This Skill Does

The stream-of-consciousness skill serves as a high-fidelity diagnostic and portability tool for the OpenClaw agent ecosystem. Its primary function is to serialize the entire conversation state—including chat history, tool executions, and system-level reasoning traces—into the standardized Open-Token format. By providing a structured, machine-parseable representation of the agent's interaction lifecycle, this skill ensures that context is never lost during complex multi-step tasks. It supports multiple output formats (JSON and NDJSON), adjustable redaction levels for PII and sensitive data, and flexible inclusion of internal system events. Whether you are debugging a failed tool call, migrating an agent between different environments, or auditing an agent's reasoning process, this skill provides the ground truth required for technical reproducibility.

Installation

To integrate this skill into your local OpenClaw environment, execute the following command in your terminal:

clawhub install openclaw/skills/skills/247arjun/stream-of-consciousness

Ensure that you have sufficient permissions to execute skills in your current project workspace. Once installed, the skill becomes immediately available as an agent tool.

Use Cases

• Agent Portability: Handoff an active conversation between two different agent runtime instances while maintaining the complete causal chain of events.
• Debugging & Auditing: Analyze complex agent behaviors by exporting internal traces and tool outputs to identify exactly where a logic loop or execution error originated.
• Data Provenance: Create immutable records of agent interactions for compliance and regulatory purposes, ensuring that all tool outputs and reasoning steps are captured.
• Model Fine-Tuning: Extract clean, structured datasets of agent reasoning to provide feedback for future model training or agent behavioral fine-tuning.

Example Prompts

1. "Export our current conversation to Open-Token format, use JSON mode with pretty printing and include full internal traces for debugging."
2. "Generate an NDJSON export of our session; apply strict redaction to any PII and limit the output to 5000 bytes."
3. "Create a conversation snapshot in summary mode to help me review the steps taken during the last data analysis task."

Tips & Limitations

• Redaction is Essential: Always specify your redact level when exporting production data to prevent sensitive information from being stored in your audit logs.
• Internal Traces: Be aware that include-internal may not return data if the host runtime environment forbids access to hidden system state; the tool is designed to handle this gracefully by setting conversation.internal_availability to unavailable.
• Truncation: Use the max_bytes parameter carefully if you are working with large conversation histories to ensure you don't lose critical start-of-conversation metadata.