@dzhechkov/scout

Multi-source ecosystem intelligence — scans 9 sources for new agent-skill projects, detects formats, scores relevance, and generates actionable intelligence reports.

Sources (9)

Usage

npm i -g @dzhechkov/harness-cli

# Quick scan — radar mode (all 9 sources)
dz scout

# Deep analysis — downloads SKILL.md, finds gaps, recommends integration
dz scout --deep

# Custom topics (GitHub-only)
dz scout --topics mcp-server,ai-agent

# Only recent
dz scout --since 2026-05-01

# Show last saved report (offline, no network)
dz scout --report

# Show what changed since last scan
dz scout --diff

What it does

Radar mode (dz scout)

1. Scans 9 sources in parallel with deduplication
2. Detects skill formats: SKILL.md, plugin.json, .claude/skills/, MCP manifests
3. Scores relevance: format (40%) + stars (30%) + recency (20%) + novelty (10%)
4. Compares against our 30 packages — finds skills we don't have
5. Reports with recommendations: integrate / monitor / skip
6. Shows source provenance — which source found each result

Deep analyst mode (dz scout --deep)

For top-scored repos (score ≥50), goes further:

1. Downloads SKILL.md from each repo, parses frontmatter
2. Finds closest match in our inventory by keyword overlap
3. Explains the delta — what the found skill adds that ours doesn't
4. Recommends integration path:
   • canonicalize — novel skill → new @dzhechkov/skills-* pack
   • merge — similar to existing → add unique features to ours
   • new-preset — novel → add to preset or new pack
   • skip — already in inventory
5. Gap analysis — trending categories we lack, with brief descriptions
6. Academic keyword extraction — arXiv/Semantic Scholar papers contribute research themes to gaps

Memory (persistent tracking)

Scout remembers repos between runs (stored at .dz/scout/):

• History tracking — every scanned repo persisted with firstSeen/lastSeen dates
• New vs seen — each scan reports how many repos are genuinely new
• Diff detection — --diff shows new/gone/score-changed repos since last scan
• Offline report — --report shows last saved report without any network calls
• User decisions — programmatically record integrate/monitor/skip for each repo

Example output

Scanning 9 sources (GitHub + npm + HN + MCP Registry + Glama + OSSInsight + Smithery + S2 + arXiv)...

Sources: github: 50, npm: 42, hackernews: 30, mcp-registry: 20, glama: 10, ossinsight: 64, smithery: 20, s2: 5

# Scout Intelligence Report
Total found: 216 | Scanned: 216

## 🟢 Integrate (high relevance)
| Repo | Stars | Formats | Score | Novel skills |
|------|-------|---------|-------|-------------|
| EXboys/evotown | 954 | agentskills-io | 88 | skill-creator, find-skills |

## 🔬 Deep Analysis (--deep only)
### EXboys/evotown (★954)
| Skill | Description | Closest match | Integration | Rationale |
|-------|------------|---------------|-------------|-----------|
| skill-creator | Create or update AgentSkills | feature-adr | **merge** | Similar — merge unique features |
| http-request | HTTP requests for API calls | — | **canonicalize** | Novel skill (954★) |

## 📊 Harness Gap Analysis
| Category | What it is | Frequency | Recommendation |
|----------|-----------|-----------|---------------|
| ai-agents | Autonomous frameworks — orchestration, planning, memory | 15 repos | Create @dzhechkov/skills-ai-agents |
| mcp | MCP servers — database, API, filesystem connectors | 10 repos | Create @dzhechkov/skills-mcp |
| tool use | LLM tool use — function calling, multi-step chains | 5 papers | Monitor — academic research trend |

Programmatic API

import { scanAllSources, deepAnalyze, generateReport } from '@dzhechkov/scout';

// Multi-source scan
const { results, totalBySource } = await scanAllSources({ token: process.env.GITHUB_TOKEN });

// Generate report
const report = generateReport({ repos: results, ... });

// Deep analysis on top results
const deep = await deepAnalyze(results, { token });

// Individual sources
import { scanNpm, scanHN, scanMcpRegistry, scanGlama } from '@dzhechkov/scout';

Status

v0.8.0 — published on npm. Part of DZ Harness Hub.