Nerissa Competitive Research
xuenerissa-png/nerissa-competitive-research
Pesquisa competitiva sem alucinações — 2 fontes obrigatórias, 6 dimensões de scoring
Ver no GitHubSkill Claude que executa pesquisa competitiva disciplinada, rejeitando atalhos preguiçosos. Exige 2+ fontes para dados de funding, verifica sobreposição de clientes, decompõe empresas em 6 dimensões e produz análise estruturada com 15 campos.
Instalação
npx convez-ai xuenerissa-png/nerissa-competitive-research1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Validar dados de funding e clientes de concorrentes antes de tomada de decisão
- Gerar relatórios estruturados de pesquisa competitiva com rastreabilidade de fontes
- Decompor estratégia de concorrentes (tech, packaging, sales, brand, network effects)
Veredito

Excelente para times que precisam rigor em competitive intelligence. A abordagem multi-fonte e dimensionalidade 6D eliminam análises rasas, mas demanda disciplina e acesso a bases de dados reais. Recomendado para startups em due diligence e product strategy.
README
A tradução para o português deste README ainda está na fila.
nerissa-competitive-research
Stop letting AI hallucinate competitor reports. A disciplined Claude Skill that refuses lazy shortcuts — every funding number must have 2 sources, every "they signed N customers" must be queryable, every "company X is similar to us" must be scored on 6 dimensions before being accepted.
The Problem
When you ask AI to research a competitor, you usually get one of these bad outputs:
- Funding numbers pulled from one secondhand blog — and the blog was wrong
- "They're an AI X startup like us" — without ever opening their website
- A description of "what they do" — never asking "what is the essence, packaging-layer or real tech?"
- A recommendation to "learn from them" — based on $400M ARR, ignoring that their customer layer doesn't match yours at all
- Inherited from an old DD report — never re-validated, so old errors propagate forever
Each of these is a real failure pattern I hit while doing competitive research for an early-stage startup. After 12 companies and many corrections, I extracted the discipline into this skill.
What This Skill Does Differently
It refuses the shortcuts. Specifically:
- Every funding claim requires ≥ 2 independent sources (Crunchbase + official PR; or LATKA + Rewardful case study)
- Every "X signed Y customers" requires querying your source-of-truth database — not memory, not last week's report
- Every company gets decomposed: technology / packaging-layer / sales motion / brand / network effect — not just "what it does"
- Every recommendation requires 6-dimension scoring — funding size alone never qualifies
- Dual-track separation: fundraising-reference (Track A · $10M+ similar direction) vs business-learning-target (Track B · path similarity, ignore funding). Mixing them causes errors.
- Inherited historical judgments must be explicitly tagged — "I didn't re-validate" is safer than "looks fine"
Quick Start
Use as a Claude Code Skill
- Clone this repo into your Claude skills directory:
mkdir -p ~/.claude/skills
cd ~/.claude/skills
git clone https://github.com/xuenerissa-png/nerissa-competitive-research.git
-
In any Claude Code session, trigger by saying:
"research [Company Name] using the competitive research skill""调研 [公司名]"(Chinese)"add [Company Name] to the competitor table — full SOP"
-
Claude will execute the 7-step SOP, score on 6 dimensions, and produce the 15-field structured output.
Use without Claude (as a manual SOP)
The SKILL.md is also a fully readable SOP document. Print it, run a research session manually, fill the 15 fields, and add to your own tracking database (Notion / Airtable / Feishu Base / etc).
A ready-to-use Feishu Base schema script is in references/feishu-base-schema.md.
Output Format
For each company researched, you get:
🏢 Company: [Name]
🔗 Website: [verified open]
💰 Funding: [≥ 2 sources cited]
🧱 Essence: [technology / packaging-layer / sales / brand / network]
👥 Customer overlap with you: [verified, not inferred]
📊 6-dimension score: X / 12
🎯 One-line "what to copy": ...
📖 3 must-see pages: ...
🚫 Anti-lazy flags caught: ...
Plus the 15-field structured record ready to drop into a database.
The 7 Steps (Summary)
- Open the official website (30s) — verify it exists, not just an aggregator blog mention
- Find funding data (5min) — Crunchbase + official PR + ≥ 2 sources
- Decompose essence (10min) — 5 questions about what it really is
- Verify customer overlap (10min) — case studies + Twitter / G2 reviews
- Score on 6 dimensions (5min) — only 8+ qualifies for Top
- Produce 3 sentences (15min) — what it is / what to copy / 3 must-see pages
- Enter structured database (5min) — 15 fields, verifiable sources
Total: 30-60 minutes per company
Examples
5 worked examples in examples/ (sanitized from real research):
-
01 · Bootstrapped SaaS via LATKA — A B2B attribution SaaS was listed in an old DD as "$500k seed." LATKA + an affiliate platform case + official "About" page cross-verified: $0 funding, fully bootstrapped, $770k ARR with 7 people. The wrong number propagated for weeks before correction.
-
02 · Funding wave packaging layer — A hot AI app-builder ($400M ARR, $6.6B valuation). Original take: "AI full-stack builder, must learn from." Essence decomposition revealed: it's a packaging layer over commodity infrastructure (UI-gen tool + BaaS + edge cloud) with a conversational interface. No new technology. Conclusion changed from "must learn" to "wide moat (scale/capital/brand) — technically anyone can copy, but they out-spent everyone on onboarding UX."
-
03 · Direct local-market competitor — Found via a tech-media report mentioning an ex-unicorn COO founding a vertical AI agent for outbound sales. Multi-source verified: founder background (international payments → unicorn distribution), tier-1 VC Pre-A funding, humanized "named AI employees" product framing. A much closer mirror to the original team's customer profile than the US enterprise giants they had been benchmarking against.
-
04 · Incumbent warning — A major platform launches a free AI agent for its ecosystem users. Not a direct competitor. But entry of a giant with free + ecosystem monetization is a structural threat to paid service models in adjacent space. Flagged for monthly monitoring, not deep research.
-
05 · Dual-track example — An enterprise AI agent giant ($1.5B funding, $15B valuation, Fortune 50 customers) vs a bootstrapped peer ($0 funding, 7 people, $770k ARR). Both relevant — but for different uses. The giant is Track A (fundraising-reference: "we're a vertical version of their outcome model"). The bootstrapped peer is Track B (business-learning: "their path is what we can actually replicate"). Confusing the two initially killed the bootstrapped peer from the recommendation list.
Why This Exists
I built this skill while doing real competitive research for an early-stage growth startup. Over a week of research, I caught myself (and Claude) making the same lazy mistakes repeatedly:
- Re-quoting funding numbers from old DD reports without re-checking
- Calling companies "competitors" without opening their website
- Recommending a hot AI app-builder as a learning target without realizing it's a packaging-layer with a moat we can't access
- Saying "we signed 3 customers" in 5 reports when source-of-truth actually showed 0 signed, 2 in quote stage, 9 in conversation
- Defaulting to a specific industry vertical based on 1 sample company, when actually only 40% of the pipeline matched that industry
Each correction pointed to a rule. This skill captures all of them.
Comparison to Other Approaches
| Approach | What it gives you | What it misses |
|---|---|---|
| One-shot AI query ("research X company") | Quick narrative | Often inherits wrong data from one source; no verification |
| Generic competitor template (Notion / Airtable) | Structure | No discipline against lazy filling; no essence decomposition |
| VC pitch deck competitor slide | Visual at-a-glance | Tends to compare on dimensions that flatter you; not honest |
| nerissa-competitive-research | All of the above + forced verification + dual-track + essence decomposition + anti-lazy checklist | More work upfront (30-60 min per company) |
If you do < 3 competitor researches a year, you probably don't need this. If you do 10+, the discipline pays back fast.
Roadmap
- v1.0 · Core SOP + 6-dimension scoring + 15-field schema + 5 examples
- v1.1 · Add Notion / Airtable schema templates (currently only Feishu Base)
- v1.2 · Add automated Crunchbase + LATKA + G2 multi-source fetcher (when reliable scrapers available)
- v1.3 · Add competitor monitoring agent (daily diff watcher for funding / pricing / customer wall changes)
Related Skills
- decision-lens — Multi-role decision review (when "should we treat X as competitor?" itself is a complex decision)
License
MIT · Free to use, fork, adapt.
Author
Built by @xuenerissa-png. If you find this useful or want to contribute a new failure pattern → rule, file an issue.
中文版 README: README_zh-CN.md