Feature Radar

Give your AI agent long-term memory and decision-making ability for feature planning.

Your AI agent is great at writing code. But ask it "what should we build next?" and it gives you generic suggestions — it doesn't remember last week's architecture decisions, doesn't know which features you already rejected, and can't connect the dots between user feedback and technical possibilities.

Feature Radar fixes this. It creates an external brain (.feature-radar/) that persists across sessions — so your agent accumulates knowledge instead of starting from zero every time.

Before / After

Before Feature Radar:

You: "What should we build next?"

Agent: "You could add dark mode, improve performance, or add user authentication..." (generic suggestions that ignore your project's context)

With Feature Radar:

You: "feature radar"

Agent: "Based on base.md, your project is a CLI tool focused on local-first workflow. I found 3 features you've already shipped that aren't documented, 2 open opportunities with high user demand, and a pattern from archive/ suggesting your streaming approach needs rethinking. I recommend building the plugin system next — it has the highest impact-to-effort ratio and aligns with your architecture."

The difference? Your agent now has context that compounds over time.

The Magic: Steer Your Agent with Annotations

The fastest way to inject your domain knowledge:

1. Say "feature radar" → agent generates opportunities
2. Open any file, add:  > NOTE: this is too complex, split into two phases
3. Say "address my notes"
4. Agent reads your notes, applies corrections, removes markers
5. Repeat until perfect

Your agent learns from your corrections. Architecture constraints, naming conventions, strategic priorities — all captured in files that persist forever.

Quick Start

With skillshare

skillshare install runkids/feature-radar --into feature-radar

With Vercel Skills CLI

npx skills add runkids/feature-radar

Manual

# Claude Code
cp -r skills/* ~/.claude/skills/

# Codex
cp -r skills/* ~/.codex/skills/

Then say "feature radar" in your next session. That's it.

You can also say "feature radar quick" for a fast scan, "feature radar evaluate" to jump to prioritization, or "feature radar #2" to deep-dive a specific opportunity.

What Happens Under the Hood

Your agent analyzes your project — language, architecture, key features — and builds a structured knowledge base:

.feature-radar/
├── base.md           ← Project dashboard: what you've built, where you're going
├── archive/          ← Shipped, rejected, or covered features (with extracted learnings)
├── opportunities/    ← Open features ranked by impact and effort
├── specs/            ← Reusable patterns and architectural decisions
└── references/       ← External inspiration, ecosystem trends, research

The data format is defined in SPEC.md — a language-agnostic specification that any AI tool can implement.

Every feature goes through a lifecycle — discovered, evaluated, built, archived. And every archived feature feeds back into the system:

flowchart TD
    subgraph Discovery
        S1["scan"] --> OPP[opportunities/]
        S2["ref"] --> REF[references/]
    end