AI support toolkit: automating product support with Claude AI, Jira integration, and LokiQL Query Generation

The problem: support in the M&S phase is every product’s bottleneck

Every digital product in the Maintenance & Support phase generates recurring support tickets. Analysts, developers, and QA teams spend hours manually reviewing logs, querying databases, reading Jira tickets, and attempting to reconstruct the context of a bug – only to discover that the exact same issue was resolved three months ago.

The result? High maintenance costs, long Mean Time to Resolution (MTTR), team frustration, and a knowledge debt that compounds with every sprint.

AI Support Toolkit is our answer to this problem: AI-driven automation that learns from historical tickets, generates diagnostic queries, and delivers a first-pass analysis of every new incident – before a human even opens Jira.

What is AI support toolkit?

AI Support Toolkit is a lightweight, modular boilerplate for automating first-line product support. It is designed as a plug-and-play solution – deployable into an existing product repository in minutes.

The toolkit is built on three pillars:

1. Claude AI Skills – a set of 12 ready-to-use skills for Claude (Claude Code) that handle every stage of ticket analysis
2. Bash scripts – automation scripts for setup, initialization, and running the tool
3. YAML knowledge base – a dynamically updated database of the most common support tickets, generated and maintained by AI

Repository: github.com/boldare/ai-support-toolkit

How it works: system architecture

1. Product codebase analysis

Before first use, the toolkit performs a deep analysis of the product’s source code. The code analysis Claude Skill identifies:

• Programming language and framework (Node.js, Python, Java, React, NestJS, etc.)
• Libraries and packages – NPM, pip, or Maven dependencies that provide context for errors
• Logging infrastructure – logger locations, logging libraries in use (Winston, Log4j, Pino), and log patterns
• System identifiers – extraction of key identifiers used throughout the application, such as userID ,companyID, sessionID, transactionID, and requestID

This phase builds a product context that feeds all subsequent Claude prompts – significantly improving the accuracy of analysis and the relevance of generated diagnostic queries.

2. Knowledge base seeding from historical Jira tickets

The toolkit pulls historical tickets from Jira and processes them through a dedicated Claude Skill for historical ticket analysis. This builds a YAML knowledge base containing:

• Common error types and their categorization
• Symptom patterns and root cause analysis
• Proven solutions and workarounds
• Related identifiers and ticket reference numbers

The knowledge base evolves automatically – every new ticket is compared against existing entries, enriching and updating the knowledge base without manual intervention.

3. New ticket analysis workflow

When a new support ticket comes in, the toolkit runs a multi-step analysis pipeline:

Step 1: Fetch the Jira ticket The Jira integration Claude Skill retrieves the full ticket data including key details: user and company identifiers, timestamps, environment (production/staging), priority, and comment history.

Step 2: Generate diagnostic queries Based on the ticket context, Claude generates ready-to-run queries:

• LokiQL – for fetching application logs from Grafana Loki, filtered by userID, time range, and error level
• SQL – for pulling relevant data from the product’s relational database

Step 3: Analysis and classification The collected logs and data are analyzed by Claude in the context of:

• Source code and known error patterns
• The historical ticket knowledge base (YAML)
• Code analysis results (identified loggers, patterns, and identifiers)

The output is a complete report with a problem diagnosis, probable root cause, and suggested resolution – before a developer takes a single manual step.

13 Claude Skills – a complete support toolkit

The heart of the boilerplate is 13 Claude Skills covering every stage of the ticket resolution process.

Workflow (ticket investigation)

Setup and maintenance

Technologies and integrations

AI Support Toolkit connects tools that already exist in most M&S team stacks:

• Claude AI / Claude Code – LLM engine for analysis and content generation
• Claude Skills (.claude/ directory) – modular AI agent skill system
• Jira REST API – source of truth for tickets, issue history, and identifiers
• Grafana Loki + LokiQL – log aggregation system with a query language for log filtering
• SQL – access to relational product data
• Bash scripting – environment setup automation and HTTP server bootstrapping
• YAML – lightweight, human-readable format for the ticket knowledge base
• HTTP Dashboard – simple web interface for browsing the knowledge base

Features and developer experience

One-command setup

The toolkit ships with setup scripts that scaffold the entire file and folder structure required by the tool. Initializing Claude Skills in a product repository is a single command – a dedicated init script configures the .claude/ directory with all 12 skills ready to use.

Knowledge base dashboard

A built-in HTML dashboard lets you browse the current ticket knowledge base without opening YAML files. A dedicated script spins up a local HTTP server with an interface for exploring, filtering, and reviewing the history of resolved tickets.

Timesheet automation

For teams tracking time in Jira, the toolkit includes an optional Skill for automatic timesheet entry generation based on completed analysis – eliminating manual time logging.

Who is AI Support Toolkit for?

Product teams in the M&S phase – especially those handling enterprise client tickets with high volumes and complex system environments.

Support engineers and L2/L3 teams – analysts looking to reduce MTTR and lower the cognitive load of ticket triage and root cause investigation.

DevOps and platform teams – engineers integrating Grafana Loki, Jira, and SQL into a unified debugging workflow.

Teams already using Claude Code – organizations that use Claude as an AI assistant in daily development work and want to extend its capabilities into the support domain.

AI-driven support: the bigger picture

AI Support Toolkit is part of the broader trend of agentic AI workflows in software engineering. Rather than treating AI as a code completion assistant, the toolkit deploys Claude as an autonomous diagnostic agent – capable of multi-step analysis, external data retrieval, and self-updating knowledge management.

This approach resembles Retrieval-Augmented Generation (RAG) patterns, where the AI model is enriched with a dynamically updated domain-specific knowledge base. In the support context, that knowledge base consists of historical tickets, and retrieval happens through the Jira API and the YAML knowledge base.

From an LLMOps perspective, the toolkit demonstrates how structured Claude Skills can replace traditional, monolithic prompts – providing better control, testability, and the ability to iterate on individual stages of the analysis pipeline independently.

Getting started

1. Clone the repository: git clone https://github.com/boldare/ai-support-toolkit
2. Run the file structure setup script
3. Initialize Claude Skills in your product repository
4. Configure Jira API credentials and your Loki/SQL connections
5. Run code analysis on your product repository
6. Seed the knowledge base with historical Jira tickets
7. Analyze your first support ticket

Full documentation and configuration examples are available in the repository README.

Summary

AI Support Toolkit solves a specific, costly problem: the time and cognitive load spent diagnosing recurring bugs in M&S products. By combining Claude AI Skills, Jira integration, LokiQL and SQL query generation, and a dynamic YAML knowledge base, the toolkit reduces first-pass ticket analysis from hours to minutes.

This is not another AI chatbot for customer support. It is an AI-powered debugging pipeline integrated directly into the tools your team already uses every day.

Check out the repository, try it on your project, and let us know how it performs in your context.