ai-radar

KO | EN

gitlite — search

#python #hacktoberfest #ai #typescript #llm #golang #go #machine-learning #linux #java #windows #pytorch #ios #open-source

(★ 12)

Claude Code skill to help you stay up to date with relevant developments

ai-radar 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# AI Radar: Claude Code Plugin

A Claude Code plugin for keeping a running, honest log of AI developments that actually matter for your project.

**What it does:** runs targeted web scans driven by your project context, scores each finding against relevance and skill-value, deep-fetches primary sources for high-relevance items, assigns explicit **✅ Yes / 🤔 Maybe / ⏭ Skip** verdicts, and appends to a retention-bounded markdown log you can review at any time.

**What it doesn't do:** fabricate entries. Every logged item is traceable to a URL fetched during the scan — no training-data recall, no confidently-wrong summaries. Restraint is a feature: scans that find nothing citable write nothing.

## Why this exists

AI tooling moves fast. Most "AI news roundups" are noise — model releases you don't care about, industry moves you won't act on, benchmarks that don't change your work. This skill spends its budget on the things *your project* names as relevant and filters everything else away. A weekly run gives you a short, scannable list of what changed and what to do about it.

## Install

From inside Claude Code, register the marketplace and install the plugin:

```bash
/plugin marketplace add https://github.com/hereinthehive/ai-radar.git
/plugin install ai-radar@gotrino
```

Verify with `/plugin list`. To update later, re-run `/plugin marketplace add` against the same URL to refresh the catalog, then reinstall.

Or clone this repo directly into your Claude Code plugin directory — see [Claude Code plugin docs](https://docs.claude.com/en/docs/claude-code/plugins).

## First-run setup

The skill lives or dies by `PROJECT_CONTEXT.md` — a short description of your project and the categories of AI developments you care about. On the first run, the skill will walk you through creating one, or you can pre-seed it:

```markdown
# Project Context

## What we're building
[2–3 sentences about the project]

## Tech stack
- [key technologies]

## What AI developments matter most

**Highest relevance — tools used daily:**
- [e.g. Claude Code, Figma, Anthropic SDK, MCP ecosystem]

**High relevance — directly usable in the product:**
- [e.g. RAG/embeddings, your framework/platform updates, relevant research]

**Medium relevance — shapes platform decisions:**
- [e.g. adjacent-but-not-immediate things]

**Low relevance — skip unless foundational:**
- [things you don't want to see]

## Preferred sources

- **Always check:** [URL or handle] — why you trust it, what it covers
- **Never cite:** [URL or handle] — e.g. an aggregator whose claims failed a primary-source check
```

The **Preferred sources** section is optional but high-leverage. When present, the skill targets those sources in its queries (via `site:<domain>` constructions or by including handles as keywords), and learns from each scan — surviving Yes/Maybe entries from new domains get offered for addition to the list.

Add sources during a scan by replying to the plan-confirmation step with plain English: *"add leerob.io to preferred sources"* or *"drop MarkTechPost — too many aggregator reposts"*.

The more specific the tooling you name, the better the scan targets. "Figma Code Connect update" produces a focused hit; "design tools" produces noise.

## Usage

Four invocation modes:

| Command | What it does |
|---|---|
| `/ai-radar` | Full scan — searches, scores, deep-fetches high-relevance items, logs, renders a summary |
| `/ai-radar show` | Render the summary of the most recent scan. No searches. |
| `/ai-radar show all` | Running "what's on my plate" digest across all scans. No searches. |
| `/ai-radar tidy` | Trim the log to the last 10 scan sections. No searches. |

Use `/loop <interval> /ai-radar` for recurring polling in a session, or `/schedule` for cron-backed remote scans.

## Design principles

- **Evidence before claims.** Entry content must trace to fetched URLs. If the fetch disconfirms a search snippet, the snippet loses.
- **Confidence over completeness.** A scan that returns four honest entries beats one that pads to seven with noise.
- **Verdict over rating.** Scores are directional; the ✅ / 🤔 / ⏭ verdict is what you actually act on.
- **Context-driven search.** The project's `PROJECT_CONTEXT.md` drives the search plan. Fall back to a generic taxonomy only when the context is thin.
- **Bounded memory.** The log keeps the last 10 scan sections — dedup still works, show-all stays scannable, context cost stays low.

## Repo layout

```
ai-radar/
├── .claude-plugin/
│   ├── plugin.json          # plugin manifest
│   └── marketplace.json     # marketplace listing
└── .claude/
    └── skills/
        └── ai-radar/
            ├── SKILL.md     # the instructions Claude Code runs
            ├── init.sh      # first-run bootstrap helper
            ├── references/
            │   └── sources.md
            └── templates/
                └── radar_template.md
```

## License

MIT — see [LICENSE](./LICENSE).

## Authored by

[Gotrino](https://gotrino.com) · Dan Donald

ai-radar

ai-radar 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

ai-radar

ai-radar 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README