Code Phantom — AI-Powered Spectral Code Guardian

“The phantom that watches over your code.”

Comment

Inspiration

We wanted to reimagine code analysis, so it feels less robotic and more intuitive. Developers often leave behind debug logs, commented code and unfinished tasks that later resurface like ghosts from another timeline. This inspired a playful, spectral assistant that observes code, warns you early, and helps you fix issues effortlessly, while keeping the experience atmospheric and delightful.

What it does

Code Phantom is a playful yet powerful code-analysis extension that watches over your code like a supernatural guardian. It reveals forgotten intent from Past You, predicts risky patterns for Future You, and offers one-click fixes to modernize and secure code. All insights surface with subtle, atmospheric magic—helpful, non-intrusive, and delightfully memorable.

Features

  • Past You - Uncovers Git history insights and detects orphaned commented code, TODO, and FIXME entries.

  • Future You - Performs real-time detection of risky patterns such as console.log, var, loose equality, deep nesting, and unsafe error handling.

  • Phantom Prophecies delivers non-blocking, randomized mystical hints with optional contextual intelligence via a custom MCP server.

  • Ghost Autocorrect offers one-click fixes: converting var → let, removing console, enforcing strict equality (===, !==), and eliminating debugger.

  • Haunting UI System visualizes severity with fading glows, whispers in gutters, cursor trails, and ethereal notifications.

  • Summoning Circle (Full Séance) generates a dramatic, actionable aggregated report with quick navigation links and fix actions.

How we built it

1. Vibe Coding with Kiro

We developed features through short, rhythmic AI conversations: intent → code → test → refine. Kiro generated analyzers for Past You and Future You, command scaffolding, diagnostics, and UI decorations. Our most impressive build was the prophecy engine delivering non-blocking, severity-aware ghost hints. Fast iteration kept requirements sharp, fun, and technically grounded.

2. Steering Docs for Response Quality

We steered Kiro with internal constraints for severity mapping, nest-depth thresholds, fade timings, and safe code zones. This improved diagnostic precision, ensured consistent ghost commentary tone, and controlled UI effect timing. Steering helped us turn AI output from exploratory prototypes into reliable feature generation. Biggest impact: accurate lint placement + consistent atmosphere.

3. Agent Hooks

Kiro generated automation hooks using VS Code debug + decoration APIs for diagnostic listening, breakpoint tracking, gutter ghosts, and warning escalation counters. Hooks automated risky pattern detection (var, consoles, loose equality) and Past-timeline Git scanning. They removed boilerplate and kept analysis active across edit and debug sessions without slowing workflows. Outcome: faster dev cycles + consistent haunted diagnostics behavior.

4. MCP (Phantom Oracle) Capability Extension

We extended Kiro with a lightweight MCP server to give deep code context awareness. MCP enabled real complexity measurement, churn density detection, deep nesting alerts, and contextual prophecies. It prioritized warnings using actual code signals, not just pattern matching. This powered next-level timeline debugging insights that would be hard to implement robustly otherwise.

Challenges we ran into

  • Making UI effects atmospheric but lightweight and performance-safe.
  • Building auto-fixes that remove lines cleanly without breaking code formatting.
  • Tuning audio effects to be spooky yet non-distracting—no lag, loops, or developer disruption.

Accomplishments that we're proud of

  • A severity-aware haunting effects system that delivers clear, non-blocking risk signals.
  • A Phantom Prophecies engine blending mystery, randomness, and real contextual analysis through MCP.
  • Fast, safe one-click Ghost Autocorrect fixes wrapped in a tool that’s both powerful and unforgettable.

What we learned

  • Goal-led vibe coding with AI speeds up complex tooling while keeping builds clear and iterative.
  • Steering conversations and Agent Hooks turn creative ideas into structured, automated workflows.
  • MCP unlocks real contextual intelligence, and audio/visual effects should amplify code signals without disrupting the editor.

Future Scope

  • Personalized prophecies based on developer habits
  • Multi-language support beyond JS/TS
  • Ambient haunted soundscapes for errors and fixes
  • Git churn heatmaps and nesting visual overlays
  • Exportable séance reports for sharing/documentation

Built With

Share this project:

Updates