Curiosity Stack
User Guide

Installation

What you need: Claude Desktop with a paid plan (Pro, Max, Team, or Enterprise). The Cowork tab must be enabled.

Your first session

When you open Curiosity Stack for the first time, you'll see a welcome message followed immediately by the scenario library — 18 pre-built topics you can start with one click.

Option A — Pick a scenario

Browse the cards by category. Click any topic that interests you. A preview panel shows all 6 layer hints. Click Start decomposition and the plugin begins immediately.

Option B — Type any topic

Skip the scenarios entirely. Just describe what you're curious about:

The plugin picks it up and starts at Layer 0 automatically. No command needed.

The 6 layers

Every decomposition follows the same structure. Understanding the layers helps you get more from each session.

Running a decomposition

Starting

Three ways to start a decomposition:

• Click a scenario card — fastest, pre-built topics with layer previews
• Type a topic — just describe what you want to understand
• Use the command — /curiosity-stack:decompose [topic]

Token-efficient flow — how v3.3.0 works

Layers L0–L6 run as plain text conversation — no HTML artifact during the session. This saves ~60% of output tokens, letting you research more topics before hitting your session limit. At the bottom of each layer you'll see: "L2 of 6 — visual artifact generates at the end."

After L6, choose from the action menu. Say "generate" to build the full visual artifact. It's generated once, with all 6 layers, in one pass.

During the session

After each layer card, you'll see a transition question. You can:

• Answer the question — adds your context to the next layer
• Say "continue" — move to next layer without adding context
• Ask a follow-up — dig deeper into the current layer before moving on
• Say "skip to L5" — jump directly to value chain actors

After L5 — India Proxy prompt

After Layer 5 is built, you'll be offered the India Proxy Agent. Say "find India proxies" to run it, or continue to L6.

After the full decomposition

You'll be offered four next actions:

• Generate the animated value chain diagram + Research Brief
• Save to your decomposition library
• Add to watchlist for weekly monitoring
• Stress test the thesis

Scenario library

18 pre-built decomposition topics across 6 categories. Designed for when you want to start immediately without having to think of a topic.

How to open it: Shows automatically on first launch. After that, type /curiosity-stack:scenarios or say "show me scenarios".

Categories

India Proxy Agent

An autonomous agent that finds Indian companies at each layer of any global value chain. In v3.3.0, the agent runs a mandatory 23-step sourcing checklist before generating output — not a keyword search, but actual page fetches across primary Indian registries and VC portfolios.

How to trigger:

• After L5 in any decomposition — say "find India proxies"
• Directly: /curiosity-stack:india-proxy [topic]
• The agent always generates an HTML artifact — never a chat summary

What the agent fetches — mandatory checklist

The agent runs three blocks before generating output. All must complete.

Block A — India registries

• NASSCOM Deep Tech — startups in NASSCOM's curated deep-tech cohorts
• DPIIT Startup India — DPIIT-recognised startups by sector
• GeM portal — government marketplace vendors (JS-rendered; searched via query)
• DSCI / CERT-In — cybersecurity domain specific
• BSE SME / NSE Emerge — SME exchange listed companies, labelled separately

Block B — VC portfolio pages (web_fetch each)

• 12 curated India VC portfolio pages — seed-stage to Series B, across deep-tech, fintech, enterprise, and consumer sectors
• + autonomous discovery of additional sector-specific VCs per run

Block C — Publications and analyst databases

• Gartner / KuppingerCole · Inc42 · Tracxn · Crunchbase India
• Domain-specific: CISO ET for cybersecurity, IISc/IIT incubation for deep-tech

Reading the sources bar

Every output shows a sources bar at the top. Each tag shows the actual fetch result:

• Green ✓ — fetched and searched successfully
• Grey "egress blocked" — page was attempted but blocked by the network. This is honest — the agent tried.
• "JS-rendered" — page couldn't be parsed (JavaScript content), searched via query instead

Map-after-fetch logic

Every company found on any portfolio page is mapped to a value chain layer after fetching — not filtered before. A GaN semiconductor company found during an EV search gets mapped to the power electronics sub-segment. No silent drops.

Output format

Three-column table: Sub-segment | Global players | India angle. Company pills show listing status: India listed (NSE/BSE main board), India listed SME (NSE Emerge / BSE SME), India private, Global, or Flagged.

Domain Knowledge

A personal company registry that the India Proxy Agent checks before any search. Use it to seed companies you know personally — pre-coverage, IISc-incubated, DRDO-adjacent, or any company that won't surface in a standard search.

How to open: /curiosity-stack:knowledge

Two ways to add companies

• Post-agent form — after every India Proxy Agent run, a "Know a company we missed?" form appears at the bottom of the output. Fill in company name, sub-segment, note. One click saves it.
• Standalone manager — /curiosity-stack:knowledge opens a full UI with sector tabs, company list, and add/edit controls.

What to seed here

• Companies you know from personal research or professional context
• IISc / IIT-incubated startups not yet in analyst databases
• DRDO licensees or spin-offs
• Companies that appeared in a VC portfolio but weren't picked up by the agent

Seeded vs Discovered labels

In every India Proxy output, companies show one of two labels:

• Seeded — you added this from your own knowledge. Signals pre-coverage, expert curation.
• Discovered — the agent found this via the sourcing checklist.

Value chain output formats

After a full decomposition, four output formats are available. Say "generate output" or choose from the post-decomposition menu.

Every output carries the ⬡ Curiosity Stack attribution badge and an invisible HTML comment for source tracking.

Thesis stress test

After building a decomposition, the stress test challenges your understanding with structured counter-arguments. Designed to surface blind spots before they become costly.

How to trigger: After a decomposition — say "stress test this" or use /curiosity-stack:stress-test

What it covers

• Bear case — what kills this thesis entirely
• Timing risk — right thesis, wrong time — what delays it
• Substitution risk — what makes this value chain irrelevant
• India-specific risks — what prevents India from capturing this opportunity
• Conviction check — 5 structured questions → research completeness score 1-10

Output is a 3-column visual: Bull case / Bear case / What to watch.

Decomposition library

Save any decomposition for future reference. Compare themes across sessions. The library builds your personal research archive over time.

How to save: After any decomposition — say "save this" or use /curiosity-stack:library save

How to retrieve: /curiosity-stack:library — shows all saved decompositions with search and filter.

Library features

• Save — stores topic, date, key insights, and value chain actors
• Search — find past decompositions by topic or keyword
• Compare — side-by-side comparison of two decompositions
• Mirror — sync library to a Google Drive folder

Watchlist + trigger alerts

Monitor any topic within its value chain on your schedule. Two modes — digest and trigger.

How to open: /curiosity-stack:watchlist

Tracker mode — regular digest

Adds a topic to your watchlist. The Watchlist Agent runs on your chosen cadence and sends a summary to Gmail.

Trigger mode — specific alerts

Sets a precise condition that fires only when something specific happens at a particular layer.

Digest format

Each digest email contains: what changed at each layer, materiality score (HIGH / MEDIUM / LOW), and a direct link to run a fresh decomposition.

Source credibility registry

Build a personal registry of trusted sources. The plugin uses this registry to weight information during research — sources you've rated highly get more prominence in outputs.

How to access: /curiosity-stack:library → Source registry tab

How it works

• Rate any source 1-5 stars with a note on its strengths and biases
• Tag sources by domain (macro, India equity, tech trends, etc.)
• The plugin auto-suggests credibility ratings when it encounters new sources
• Set auto-accept to let the plugin add sources automatically

All commands

You rarely need these — most features activate through conversation. But if you want direct access:

Test checklist

Use this to verify every feature is working after installation or after an update.

Core features

• Welcome + scenario library — Open new conversation. Scenario cards should appear automatically without typing anything.
• Scenario launch — Click any scenario card → preview → Start decomposition. Layer 0 should begin immediately.
• Free-text topic — Type "Help me understand grid-scale battery storage". Layer 0 should start without a command.
• Full decomposition — Run a full session through all 6 layers. Each layer should produce a card. Time: 8-12 minutes.
• Value chain output — After L6, say "generate output". Value Chain diagram should render with animations.
• Research Brief — Say "generate Research Brief". Should produce a clean one-page forwardable document.

Agent features

• India Proxy Agent — After L5, say "find India proxies". Agent should search autonomously and return a structured shortlist.
• Watchlist add — Say "add [topic] to my watchlist, weekly". Should confirm addition and cadence.
• Trigger alert — Say "alert me when [specific condition]". Should confirm trigger is set.
• Watchlist agent — Run /curiosity-stack:watchlist. Should show current watchlist items.

Library features

• Save decomposition — After a session, say "save this". Should confirm save with topic and date.
• Retrieve library — Run /curiosity-stack:library. Should show saved decompositions.
• Stress test — After a decomposition, say "stress test this". Should return 3-column bull/bear/watch output.

SEBI compliance

• Disclaimer on outputs — Every generated output should carry the SEBI disclaimer at the bottom.
• No buy/sell language — The plugin should never say "buy", "sell", "invest in", or "recommended stock".
• Company framing — Companies at L5 should be described as "operating at this layer" — not as recommendations.

Tips and tricks

Getting better outputs

• Be specific at L0 — "What's happening with India's semiconductor push in 2026" gets better results than "semiconductors"
• Add your context — At any layer, say "I already know X, focus on Y" to skip what you know
• Use the transition questions — Answering each layer's transition question adds your context and makes the next layer more relevant
• Run setup — Setting your geography and themes in setup makes India proxies and geographic context automatic

Workflow patterns

• Quick scan — Skip to L5 directly: "Jump to value chain actors for [topic]"
• Deep dive — Run the full 6 layers + India proxy + stress test in one session
• Weekly review — Add 3-5 topics to watchlist, let the digest surface what's changed
• Before a meeting — Run a quick decomposition on the company/sector being discussed. The Research Brief format is designed for this.

Sharing outputs

• Every output has a share as card option — generates a LinkedIn/X formatted image
• Research Brief is designed to be forwarded — use it in emails or Slack
• The ⬡ Curiosity Stack badge on all outputs builds brand recognition if you share publicly
• Use #CuriosityStack when sharing on X

Troubleshooting