VC Signals

Turn noisy internet chatter into ranked investor briefs.
One command per week. 8 emerging themes with company mapping.

Claude Code Co-Work Compatible Works Out of the Box

00 Get Started

Three install methods. The skill walks you through setup on first run.

Plugin Install (easiest)

Inside Claude Code or Co-Work, type:

/plugin marketplace add https://github.com/abhishek255/vc-signals

Terminal Install

One command in your terminal:

curl -sL https://raw.githubusercontent.com/abhishek255/vc-signals/main/install.sh | bash

Clone the Repo

For developers who want to inspect or contribute:

git clone https://github.com/abhishek255/vc-signals.git cd vc-signals && claude

First Run: Auto-Setup

On your first command, the skill detects it's new and offers a guided 2-minute setup. Claude installs the research engine, walks you through API keys one at a time (paste or skip), and saves your config. Skip setup entirely to get results immediately via web search.

01 Architecture

Three layers. Claude is the brain. Python scripts handle API calls. SKILL.md ties it together.

You type a command

/vc-signals weekly devtools

SKILL.md (Orchestrator)

Parse args, load config, select retrieval path

Retrieval

WebSearch (default)
or last30days (Reddit, HN, X, YouTube)

GitHub Trending

Star velocity API
Fast-growing repos

Claude's Brain

Extract themes, cluster, score momentum, map companies, frame for investors

Investor Brief

Printed inline

Saved Locally

JSON + Markdown for week-over-week tracking

02 Five Modes

Each mode serves a different part of the VC research workflow.

/vc-signals weekly devtools

Weekly Sector Scan

Top 8-12 emerging themes. Ranked by momentum with company mapping and hype-vs-durable verdicts.

### 1. MCP Protocol & Agent Infra Momentum: 9/10 | Confidence: High Timing: Early-Mid Companies: MintMCP, Anthropic, Perforce
/vc-signals theme "agent evals"

Theme Drill-Down

Deep analysis with Perplexity synthesis (50+ citations) when available. Why now, subthemes, companies, OSS projects, durable vs hype.

## Why Now? As AI agents move from demos to production, evaluation frameworks became the critical gap...
/vc-signals company "Confluent"

Company Backtrace

Theme exposure mapping with GitHub repo/founder search and X/Twitter handle targeting. Role classification, evidence, competitive context.

Theme Exposure: Streaming Infra | Direct solver AI Data Pipeline | Beneficiary Event Architecture | Adjacent
/vc-signals github ai-infra

GitHub Trending

Top repos by star velocity. Maps fast-growing OSS projects to commercial entities.

# | Repo | 7d | Stars 1 | macos-iso-build | +116 | 1,041 2 | claude-security | +82 | 4,198 3 | agents-radar | +39 | 613
/vc-signals setup

Setup Wizard

Guided walkthrough for API keys. Claude clones the research engine, walks you through each key, saves config. Runs automatically on first use.

/vc-signals add-sector fintech

Add Sector

Claude proposes subcategories, generates search queries, subreddits, and filters. Saves to config. No JSON editing.

03 Retrieval Pipeline

Two paths, selected automatically. WebSearch works day one. Setup unlocks source-level diversity.

Zero-Config Path

Uses Claude's built-in WebSearch. 8-12 targeted queries per sector. Works immediately, no API keys needed.

What you get: Broad web coverage, blog posts, news articles, some forum content.

What you miss: Per-source engagement data, subreddit targeting, X/Twitter discussions.

Enhanced Path (after setup)

Uses last30days engine with auto-resolve — automatically discovers relevant subreddits, X handles, and GitHub context for each query. No hardcoded lists.

What you get: Source-level diversity, engagement data, auto-discovered subreddits/handles, GitHub person/project search, X handle targeting.

Deep research: Theme drill-downs use Perplexity Sonar Pro (50+ citations) when OpenRouter key is configured.

Data Sources

🌐
Web Search
Built-in
📰
Hacker News
After setup (free)
💬
Reddit
Auto-resolve discovers subreddits
🐙
GitHub
PAT or gh CLI, person/project mode
🐦
X / Twitter
Browser tokens, handle targeting
🎥
YouTube
ScrapeCreators
🔎
Perplexity
OpenRouter key, deep research

● Green border = works after basic setup    ● Orange border = needs extra credentials (optional)

04 Momentum Scoring

Each theme gets a 1-10 score. Transparent and explainable — Claude justifies every number.

Scoring Factors

Recency
High
Source Diversity
High
Repetition Density
Medium
Engagement Signals
Medium
Novelty
Medium
Technical Specificity
Low
GitHub Velocity
Low
8-10
Breakout
Multiple sources, very recent, high engagement, new conversation
5-7
Rising
Clear signal but fewer sources, or overlaps with known trends
3-4
Ambient
Mentioned but not clearly accelerating, could be noise
1-2
Faint
Single source, low engagement, possibly stale

05 Company Mapping

Three layers from most to least reliable. Every mapping gets a confidence tag.

1

Curated Seed Map

40 hand-maintained entries in company_aliases.json. Known companies, OSS projects, sectors, themes. Users can edit.

2

Evidence-Based Extraction

Claude identifies companies mentioned in search results. Captures context, source URL, and role classification.

3

GitHub to Commercial Entity

Maps trending repos to companies via org ownership, README links, pricing pages.

Confidence Tags

Confirmed
In seed map or multiple sources corroborate
Likely
Strong contextual evidence, one clear source
Inferred
Claude's judgment on limited evidence
Speculative
Thin signal, flagged as uncertain

Role Classification

Rising Theme
e.g., AI Code Review
Direct Solver
CodeRabbit, Qodo
Beneficiary
GitHub, GitLab
Adjacent
Datadog, Snyk

06 Week-over-Week Tracking

Every scan is saved. Future runs compare against previous scans to surface what's changing.

Week 1
First scan — baseline. 8 themes identified.
Week 2
2 new themes, 1 fading, "AI Code Review" momentum 6 → 8.
Week 3
"MCP Protocol" appears 3rd consecutive week → marked Durable.
Week 4
"WebAssembly Runtimes" dropped out. 1 new breakout signal.

What Gets Compared

New
Themes that weren't in last scan
Accelerating
Momentum gained +2 or more
Fading
Dropped out or lost 3+ momentum
Durable
3+ consecutive weeks → real signal

Theme Index

"AI-Powered Code Review": { "first_seen": "2026-03-15", "last_seen": "2026-04-10", "appearances": 4, "momentum_history": [6, 7, 7, 8], "peak_momentum": 8 }

07 Sector Coverage

Three sectors, six subcategories each. Editable via sectors.json. Add your own sectors.

devtools

  • CI/CD & Build Systems
  • Testing & Quality
  • IDEs & Developer Experience
  • Observability & Debugging
  • Infrastructure as Code
  • Source Control & Code Mgmt

cybersecurity

  • Application Security
  • Cloud & Infra Security
  • Identity & Access
  • SOC & Detection
  • AI & LLM Security
  • Data Security & Privacy

ai-infra

  • Model Serving & Inference
  • Training & Compute
  • Data & Feature Pipelines
  • Agent Infrastructure
  • MLOps & Experiment Mgmt
  • AI-Native Dev Tools

08 Graceful Degradation

The skill never crashes. Every failure has a fallback.

What Can Fail

  • last30days not configured
  • GitHub rate limited
  • GitHub token missing
  • File system not writable
  • WebSearch returns thin results
  • requests library not installed

What Happens Instead

  • Falls back to WebSearch (still works)
  • Returns partial results + warning
  • Skips GitHub trending, suggests /setup
  • Prints brief inline, skips save
  • Notes limited coverage, lowers confidence
  • Skips GitHub trending, returns clean JSON