persona-inspect

Diagnose persona behavior by inspecting SQLite data and LLM trace records.

34|12|Updated Aug 1, 2021
One-click install
npx skills add https://github.com/pear-studio/nonebot-dicepp --skill persona-inspect
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: persona-inspect
Source: https://github.com/pear-studio/nonebot-dicepp/tree/main/docs/agent/skills/persona-inspect
Command: npx skills add https://github.com/pear-studio/nonebot-dicepp --skill persona-inspect

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

When persona-related features behave unexpectedly, it’s hard to pinpoint whether the underlying SQLite data—especially cross-table aggregates and LLM trace records—is correct.

Core Features & Use Cases

  • Cross-table persona aggregation inspection: quickly view user persona “whole picture” including relationships, recent messages, score changes, and today’s usage.
  • LLM health and trace troubleshooting: inspect error distributions, latency percentiles, max_rounds, daily usage, and format round-level think/tool_call/tool_result details for the trace pipeline.
  • Schema discovery (DDL): list all persona_ prefixed tables’ DDL to support targeted diagnosis and verification.

Quick Start

Run the persona-inspect tool to fetch a user’s persona profile in the target bot database by executing a single command like: python scripts/dev/persona_inspect.py user <user_id> --bot-id <bot_id>.

Frequently Asked Questions about persona-inspect

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I debug incorrect persona behavior stored in a SQLite database?

Debug incorrect persona behavior by inspecting aggregated SQLite persona data and LLM trace records using a single CLI command. The tool performs read-only diagnostics across multiple persona_ tables to pinpoint cross-table aggregation issues and relationship signals.

How can I inspect LLM latency metrics and error distributions for a specific bot?

Inspect LLM latency metrics and error distributions by running the trace subcommand on the target SQLite database. It formats round-level think, tool_call, and tool_result details while displaying max_rounds, daily usage, and latency percentiles for the specified bot.

Can I view the DDL schema of persona tables without modifying the SQLite database?

Yes, you can list DDL schemas for all persona_ prefixed tables without modifying the database. The tool requires only read-only access via the --db and --bot-id parameters, ensuring safe schema discovery and targeted diagnosis during troubleshooting.

Why does a user's persona score change unexpectedly during interactions?

Persona score changes can be diagnosed by inspecting cross-table persona aggregations, recent message history, and relationship signals. The tool aggregates user persona state from the SQLite database to reveal score modifications and today's usage patterns for the target bot.

What is the best way to trace LLM call chains for bot diagnostics?

The best way to trace LLM call chains is using the dedicated trace inspection subcommand on the SQLite database. It formats the round-level trace pipeline details, exposing LLM health metrics and tool call sequences to isolate failures within a specific bot's execution flow.