code-analysis

Example: Code Analysis

Analyze a sample Python project for bugs, security issues, and quality improvements. Uses CodeAnalysisPrompt — a TOON-aware prompt builder that understands the compact notation and produces structured findings with file/line references.

Quick Start (Python — 1 line)

from toonic.server.quick import run
run("./examples/code-analysis/sample-project/", goal="find bugs, security issues, suggest improvements", interval=0)

Quick Start (fluent builder)

from toonic.server.quick import watch

server = (
    watch()
    .code("./examples/code-analysis/sample-project/")
    .goal("find bugs, security issues, suggest improvements")
    .interval(0)  # one-shot
    .build()
)

Quick Start (CLI)

python -m toonic.server \
  --source file:./examples/code-analysis/sample-project/ \
  --goal "find bugs, security issues, and suggest improvements" \
  --interval 0

Continuous Analysis (watch mode)

# Re-analyze every 60s — detects file changes via delta TOON
python -m toonic.server \
  --source file:./examples/code-analysis/sample-project/ \
  --goal "code review: find bugs, security issues, suggest improvements" \
  --interval 60

# Combine with log monitoring for full-stack analysis
python -m toonic.server \
  --source file:./examples/code-analysis/sample-project/ \
  --source log:./app.log \
  --goal "correlate code issues with runtime errors" \
  --interval 30

Using CLI Shell

python -m toonic.server.client
toonic> status
toonic> analyze find all security vulnerabilities
toonic> convert examples/code-analysis/sample-project/main.py toon
toonic> history 10
toonic> query "findings with high severity"
toonic> sql SELECT target_path, content FROM exchanges WHERE action_type='code_fix'

How It Works

1. FileWatcher scans sample-project/ and converts to TOON format
2. LLM Pipeline auto-selects CodeAnalysisPrompt (detects code category + goal keywords)
3. CodeAnalysisPrompt builds TOON-aware prompt: explains M/c/f/m/i/e notation to LLM
4. LLM returns structured JSON with findings[] — file, line, severity, description, fix
5. ResponseParser extracts JSON from markdown fences or raw response
6. Results stored in toonic_data/history.db and shown in Web UI

Prompt Builder Selection

The select_prompt_builder() function auto-selects based on goal + data categories:

Condition	Builder	Optimized for
Video sources + CCTV keywords	CCTVEventPrompt	Event analysis
Code/config sources or code keywords	CodeAnalysisPrompt	TOON-aware code review
Everything else	GenericPrompt	General analysis

Expected Findings

The sample project has intentional issues:

• Bug: deactivate_user() — no existence check, KeyError on missing user
• Bug: create_order() — no validation on total, could be negative
• Security: process_payment() — hardcoded API key
• Performance: get_orders() — O(n) linear scan instead of indexed lookup

Files in This Example

• README.md — this file
• sample-project/main.py — Python code with intentional bugs
• sample-project/config.py — Config file with hardcoded secrets