skills.rest
haJ1t
haJ1tCOMMUNITY
Explore all skills by this author →
GitHub

agent-introspector

Inspect AI agent decision traces, tool calls, context, and token usage.

Updated May 24, 2026
Search Tags:#performance profiling#AI debugging#decision trace#tool call audit#agent introspection#context inspection
One-click install
npx skills add https://github.com/haJ1t/senior-dev-squad-skills --skill agent-introspector
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: agent-introspector
Source: https://github.com/haJ1t/senior-dev-squad-skills/tree/main/plugins/agent-platform-pro/skills/agent-introspector
Command: npx skills add https://github.com/haJ1t/senior-dev-squad-skills --skill agent-introspector

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve?

The agent-introspector Skill addresses the problem of understanding and diagnosing AI agent behavior and reasoning, ensuring reliability and correctness in automated processes.

Core Features & Use Cases

  • Decision Trace Analysis: Review the reasoning chain and decision-making process of an AI agent.
  • Tool Call Audit: Inspect tool calls for redundancy, inefficiency, and failures.
  • Context Window Inspection: Verify the context state, identifying missing or compacted information.
  • Token Usage Breakdown: Analyze token consumption across phases, tools, and files.
  • Hallucination Detection: Ensure every claim made by the agent is verified with sources.
  • Performance Profiling: Measure the agent's performance and identify bottlenecks.
  • Use Case: If the AI agent provides incorrect output or performs unexpectedly, use the agent-introspector to trace its decision-making process and identify the source of the error.

Quick Start

Start introspection by querying the decision chain of the latest decision made by the agent.

Frequently Asked Questions about agent-introspector

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

FAQPage Schema
How do I debug an AI agent that provides incorrect output or performs unexpectedly?▼

Agent introspection debugs incorrect AI output by tracing the decision-making process, analyzing the reasoning chain, and inspecting tool calls to pinpoint the exact source of the error.

What is agent introspection for validating automated processes?▼

Agent introspection is the technique of examining an AI agent's internal reasoning, context state, and tool calls to ensure reliability and correctness in automated processes by validating outputs and detecting hallucinations.

Can I audit tool calls for redundancy and inefficiency in my AI agent?▼

Yes, you can perform a tool call audit to inspect AI agent tool calls, identifying redundancy, inefficiencies, and failures within the automated workflow to improve overall performance.

How do I analyze token usage and performance bottlenecks in an AI agent?▼

You analyze token usage and performance bottlenecks by running performance profiling, which measures the AI agent's performance and breaks down token consumption across different phases, tools, and files.

How do I inspect the context window to identify missing or compacted information?▼

Context window inspection verifies the AI agent's context state, allowing you to identify missing or compacted information, ensuring the agent operates with the complete required context for accurate decisions.

What is the best way to detect hallucinations in AI agent reasoning?▼

The best way to detect hallucinations is using hallucination detection during introspection, which ensures every claim made by the AI agent is verified with sources, confirming the reliability of the decision trace.

Related Skills

Explore More Software Engineering Skills→
alangebalangeb

think

Analyze session audit logs to detect execution loops and adjust strategies.

Community
Advanced
GamezarGamezar

customaize-agent-context-engineering

Explain context components, attention mechanics, and degradation patterns in agent systems.

Community
Advanced
saeed-vayghansaeed-vayghan
30

error-detective

Correlate logs, traces, and metrics to diagnose root causes in distributed systems.

Community
Advanced
itsadijmbtitsadijmbt
1

dingo-data-quality

Evaluate datasets and generated content with configurable rule-based and LLM evaluators.

Community
Advanced