session-logs

The session-logs skill provides a robust interface for querying and analyzing the historical activity of the GeminiClaw agent framework. It is designed for developers and power users who need to introspect into multi-turn conversations, audit agent decision-making, and perform cost analysis across autonomous operations. By utilizing structured JSONL data stored in the workspace memory, this skill enables programmatic access to response text, tool execution status, and token consumption metrics, making it an essential tool for performance optimization and debugging complex agent workflows.

• Perform granular keyword searches across all historical session responses using ripgrep (rg) for high-speed pattern matching.

• Analyze tool execution frequency and identify problematic tool calls that resulted in errors or unexpected failures.

• Calculate cumulative and daily token consumption trends to manage budget and optimize LLM prompt efficiency.

• Retrieve structured session data, including RunIDs, timestamps, and specific trigger sources like manual input or heartbeat events.

• Access and search Obsidian-compatible markdown summaries that capture TL;DR, key decisions, and conversation context.

• Leverage jq filters to parse nested JSONL structures for specific metadata, such as failed tool calls or specific tool names like browser_navigate.

• Ensure the {workspace} path is correctly set as per the geminiclaw_status tool or your GEMINI.md configuration.

• Note that summaries are automatically generated when a session remains idle, but can be forced manually via the geminiclaw session summary command.

• Use jq to filter out null entries if you encounter file processing errors with the first line of JSONL logs.

• The skill is intended to run within the GeminiClaw environment, ensuring that analysis remains localized within the Docker sandbox for maximum security and privacy.

• For optimal utility, ensure you are running these queries within a terminal that has access to the standard unix toolset including jq, rg, and awk.