Docs/Getting started

Practical Examples

Real-world examples showing how to use TachiBot tools and workflows effectively. From simple tool calls to complex multi-step orchestrations.

Using Tools Directly (No Workflows)

You can call any TachiBot tool directly from Claude Code without creating a workflow. Just use natural language:

Example 1: Simple Research

Ask Claude to use perplexity_ask tool:

You

"Use perplexity_ask to research the latest developments in Rust async runtime"

Claude

I'll use the perplexity_ask tool to research that.

Tool call

perplexity_ask({
  "query": "latest developments in Rust async runtime 2024",
  "searchDomain": "general"
})

Result

Shows recent tokio updates, async-std developments, and more.

Example 2: Code Analysis

Direct code review with qwen_coder:

You

"Use qwen_coder to review this code for bugs:"

function validateEmail(email) {
  return email.includes('@');
}

Claude

I'll use qwen_coder to review this code.

Tool call

qwen_coder({
  "task": "review",
  "requirements": "Check for bugs and security issues",
  "code": "function validateEmail(email) { return email.includes('@'); }",
  "language": "javascript"
})

Result

Found issues: 1. Weak validation — doesn't check domain 2. Missing null checks 3. No email format validation (RFC 5322) Recommendation: Use proper regex or validation library

Multi-Tool Chains

Chain multiple tools together for more powerful analysis. Claude can orchestrate this automatically:

Research → Validate → Analyze → Save

Complete analysis pipeline in natural language:

You

"Research GraphQL vs REST with perplexity, validate findings with grok_search, analyze the trade-offs with qwen_coder, and save the results to a markdown file with timestamp"

Claude

Claude executes a 4-step pipeline: 1. perplexity_reason({ query: "GraphQL vs REST API comparison" }) 2. grok_search({ query: "GraphQL REST performance benchmarks 2024" }) 3. qwen_coder({ task: "analyze", requirements: "Compare based on research" }) 4. Saves output to timestamped markdown file

Result

File created: ./graphql-vs-rest-2025-11-14-16-30.md

Multi-Model Consensus

Get opinions from 3 different models:

You

"Should I use TypeScript or Go for a new microservice API? Get recommendations from grok_reason, gemini_brainstorm, and openai_brainstorm"

Claude

I'll gather perspectives from multiple models. 1. grok_reason says: "Go for performance-critical services..." 2. gemini_brainstorm suggests: "TypeScript for faster development..." 3. openai_brainstorm recommends: "Consider team expertise..."

Result

Based on 3 model consensus: • Go: Better for CPU-intensive, low-latency services • TypeScript: Better for rapid prototyping, Node.js ecosystem Recommendation: TypeScript for MVP, migrate to Go if performance becomes bottleneck

Running Workflows

Workflows are pre-built multi-step processes. TachiBot includes 11 ready-to-use workflows:

How to Run a Workflow

terminalshell

1# Basic syntax
2workflow --name {workflow-name} --query "{your input}"
3 
4# Example: Code review
5workflow --name code-review --query "$(cat src/app.js)"
6 
7# Example: Creative brainstorming
8workflow --name ultra-creative-brainstorm --query "AI-powered developer tools"
9 
10# Example: Run from custom file
11workflow --file ./my-workflow.yaml --query "analyze this codebase"

Available Workflows

ultra-creative-brainstorm18 steps, 14 techniques

Comprehensive creative ideation pipeline

code-review7 steps, parallel execution

Multi-perspective code analysis

iterative-problem-solver15 steps, conditional

Research → Analyze → Solve with refinement

code-architecture-review7 steps

Systematic architecture analysis

accessibility-code-auditMultiple steps

WCAG compliance checking

pingpong8 steps, debate

Multi-model debate and refinement

Multi-Model Verification & Critique Tools

For consensus verification and critical analysis, use these MCP tools directly (no workflow file needed):

jury

Multi-model jury panel + Gemini judge synthesis for consensus verdicts

plan_critique

Adversarial red-team of an existing plan: pre-mortem, hidden assumptions, ranked risks

diff_review

Multi-model diff-aware code review, deduplicated and severity-ranked by a Gemini judge

Finding & Reading Workflow Outputs

Output Directory Structure

All workflow outputs are saved to workflow-output/ directory:

workflow-output/shell

1workflow-output/
2├── ultra-creative-brainstorm/
3│   └── 2025-11-14-16-30-abc123/
4│       ├── manifest.json                                    # Execution metadata
5│       ├── 1-research-context-gemini-3-1-pro-2025-11-14-16-30-45-Thursday.md
6│       ├── 2-analyze-patterns-grok-4-5-2025-11-14-16-31-12-Thursday.md
7│       ├── 3-brainstorm-ideas-gpt-5-6-sol-2025-11-14-16-32-05-Thursday.md
8│       └── ...
9├── code-review/
10│   └── 2025-11-14-17-00-def456/
11│       ├── manifest.json
12│       └── ...
13└── ...

File Naming Convention

{stepNumber}-{stepName}-{modelName}-{YYYY-MM-DD-HH-MM-SS-DayName}.md

Example: 3-brainstorm-ideas-gpt-5-6-sol-2025-11-14-16-32-05-Thursday.md

Configuration

Override default output directory in your .env file:

.envenv

1# .env file
2WORKFLOW_OUTPUT_DIR=./my-custom-outputs
3 
4# Or use absolute path
5WORKFLOW_OUTPUT_DIR=/Users/you/Documents/tachibot-results

Reading Results

1. Check Manifest.json — contains execution summary and metadata:

manifest.jsonjson

1{
2  "workflowId": "abc123",
3  "workflowName": "code-review",
4  "status": "completed",
5  "startTime": "2025-11-14T16:30:00Z",
6  "endTime": "2025-11-14T16:35:00Z",
7  "query": "Review authentication module",
8  "steps": [
9    {
10      "id": "1",
11      "name": "security-scan",
12      "status": "completed",
13      "output": "Found 2 potential vulnerabilities...",
14      "file": "1-security-scan-grok-4-5-2025-11-14-16-30-45-Thursday.md"
15    }
16  ]
17}

2. Read Individual Step Outputs

terminalshell

1# Navigate to workflow output
2cd workflow-output/code-review/2025-11-14-16-30-abc123/
3 
4# Read step-by-step
5cat 1-security-scan-grok-4-5-2025-11-14-16-30-45-Thursday.md
6cat 2-code-quality-gemini-3-1-pro-2025-11-14-16-31-12-Thursday.md
7 
8# Or view all outputs
9ls -lh *.md

3. Debugging Failed Workflows — use jq to find errors in manifest.json:

terminalshell

1# Find failed steps
2jq '.steps[] | select(.status == "failed")' manifest.json
3 
4# Get error details
5jq '.steps[] | select(.error != null) | {step: .id, name: .name, error: .error}' manifest.json
6 
7# Output example:
8{
9  "step": "3",
10  "name": "brainstorm-ideas",
11  "error": "API rate limit exceeded"
12}

Real-World Use Cases

Use Case 1: Analyzing a GitHub Issue

Multi-tool analysis pipeline:

You

"Analyze GitHub issue #234 about performance regression: 1. Research similar issues with perplexity_reason 2. Analyze the code changes with qwen_coder 3. Get optimization suggestions from grok_code 4. Validate solution with the jury tool"

Result

Complete analysis with: • 5 similar historical issues found • Root cause identified (N+1 query problem) • 3 optimization strategies suggested • Solution verified by 3 models (95% confidence)

Use Case 2: Technical Decision Making

Compare database options:

You

"Run iterative-problem-solver workflow to decide between PostgreSQL and MongoDB for a social media analytics platform with 1M daily active users"

Claude

Workflow executes 15 steps: Step 1–3: Research both databases (perplexity + grok) Step 4–6: Analyze requirements vs capabilities Step 7–10: Challenge assumptions with different scenarios Step 11–13: Verify recommendations with multi-model consensus Step 14–15: Synthesize final decision with trade-offs

Result

Output: 15-page analysis in workflow-output/ Recommendation: PostgreSQL for ACID compliance, with MongoDB for time-series data

Use Case 3: Code Refactoring Plan

Comprehensive refactoring strategy:

You

"Use code-architecture-review workflow on src/ directory"

Claude

Workflow analyzes: • Code structure and patterns • Coupling and cohesion • Performance bottlenecks • Security vulnerabilities • Test coverage gaps

Result

Delivers: ✓ 12 refactoring recommendations ✓ Prioritized by impact (high/medium/low) ✓ Before/after code examples ✓ Estimated effort (story points) ✓ Risk assessment for each change Saved to: workflow-output/code-architecture-review/2025-11-14.../

Advanced Patterns

Parallel Execution

Run multiple tools simultaneously for faster results:

yamlyaml

1# In your workflow YAML
2- name: gather-perspectives
3  parallel: true
4  tools:
5    - name: perplexity-view
6      tool: perplexity_ask
7      input: { query: "${query}" }
8 
9    - name: grok-view
10      tool: grok_reason
11      input: { problem: "${query}" }
12 
13    - name: gemini-view
14      tool: gemini_brainstorm
15      input: { prompt: "${query}" }
16 
17# All 3 tools run at the same time!

Conditional Execution

Execute steps based on previous results:

yamlyaml

1- name: validate-solution
2  tool: gemini_judge
3  input:
4    perspectives: "${solution}"
5    mode: "evaluate"
6  output: { variable: validation }
7 
8- name: refine-if-needed
9  tool: kimi_thinking
10  input:
11    problem: "Improve solution based on feedback"
12  when: "${validation.confidence} < 0.8"  # Only runs if confidence low

File-Based Chaining (Large Outputs)

Handle outputs larger than 1MB by saving to disk:

yamlyaml

1- name: deep-research
2  tool: perplexity_reason  # High maxTokens, generates large output
3  saveToFile: true     # Automatically saves to disk
4  maxTokens: 90000     # Can use very high limits
5  output: { variable: research }
6 
7- name: analyze-findings
8  loadFiles: ["deep-research"]  # Loads from disk, not memory
9  tool: gemini_analyze_text
10  input:
11    text: "${research}"  # Full 90k tokens available!

Prompt Engineering Integration

Use 31 research-backed techniques for better reasoning:

yamlyaml

1- name: decompose
2  tool: think
3  promptTechnique: problem_decomposition
4  input: { thought: "Break down ${query}" }
5 
6- name: ideate
7  tool: gemini_brainstorm
8  promptTechnique: what_if_speculation
9  input: { prompt: "Generate ideas for ${query}" }
10 
11- name: challenge
12  tool: grok_reason
13  promptTechnique: first_principles
14  input: { problem: "Validate ${ideas}" }
15 
16# Available techniques: problem_decomposition, what_if_speculation,
17# first_principles, pattern_recognition, and 10 more!

All Tools Reference →Complete documentation for all 31 tools Workflows Guide →Learn to create custom workflows Tool Profiles →Pre-configured tool sets Tools Configuration →Customize which tools are enabled

✓

Pro Tip

Start with direct tool calls to understand what each tool does. Once you find a pattern you use often, convert it to a workflow for repeatability. Use the create_workflow tool to generate YAML from your natural language description!

Loading documentation...

← Previous

Quick Start

Tools Overview

Edit this page on GitHub

Using Tools Directly (No Workflows)

Example 1: Simple Research

Example 2: Code Analysis

Multi-Tool Chains

Research → Validate → Analyze → Save

Multi-Model Consensus

Running Workflows

How to Run a Workflow

Available Workflows

Multi-Model Verification & Critique Tools

Finding & Reading Workflow Outputs

Output Directory Structure

Configuration

Reading Results

Real-World Use Cases

Use Case 1: Analyzing a GitHub Issue

Use Case 2: Technical Decision Making

Use Case 3: Code Refactoring Plan

Advanced Patterns

Parallel Execution

Conditional Execution

File-Based Chaining (Large Outputs)

Prompt Engineering Integration

Related Resources

Using Tools Directly (No Workflows)

Example 1: Simple Research

Example 2: Code Analysis

Multi-Tool Chains

Research → Validate → Analyze → Save

Multi-Model Consensus

Running Workflows

How to Run a Workflow

Available Workflows

Multi-Model Verification & Critique Tools

Finding & Reading Workflow Outputs

Output Directory Structure

Configuration

Reading Results

Real-World Use Cases

Use Case 1: Analyzing a GitHub Issue

Use Case 2: Technical Decision Making

Use Case 3: Code Refactoring Plan

Advanced Patterns

Parallel Execution

Conditional Execution

File-Based Chaining (Large Outputs)

Prompt Engineering Integration

Related Resources