/vc-research

name: vc-research
description: Find startup opportunities, connect people and projects, and prepare focused VC follow-ups without turning discovery into diligence.

VC research

This workflow supplies domain judgment, not an execution budget. Respect the selected light/deep mode, available tools and source scope. Use the shared research skill for discovery mechanics and durable notes when needed; do not duplicate its bookkeeping or add another completion gate.

• Identify the current job: opportunity discovery by default; meeting preparation

or diligence only when requested. Do not automatically run every stage. Infer useful starting directions without inventing stage, geography, funding, check size or traction requirements. Follow-ups address the new request and affected findings, not a restart of completed work.

• Calibrate novelty for an experienced VC before recommending: assess whether

the direction and projects are already widely discussed using observed dates, repeated coverage and ecosystem activity. Familiar or already somewhat hot projects are brief background/comparables, not the main recommendations. Seek less-discussed builders, emerging wedges and non-obvious implications. Low visibility alone is not merit; explain the substantive signal. Popularity is a provisional assessment, not a hard filter or an exhaustive survey.

• When sources are automatic and the request leaves platforms unspecified or

the scope broad, default to Xiaohongshu, X, WeChat public articles, podcasts and GitHub. Explicit user platforms always take precedence. For discovery, delegate complementary platform questions together with explore_platforms; synthesize an opportunity map, then send focused follow-up questions with known findings and gaps in a later round. Do not restart all five on a narrow follow-up. Note unavailable sources; never claim coverage from failed workers.

• For sourcing, find people worth learning about: demos, repositories, early

users, hiring, cofounder requests, unusual approaches and practitioner discussions. Explore complementary directions and follow unexpected relevant signals. Unknown identity, funding or revenue is a label, not an admission gate. Keep useful small and unnamed projects alongside the shortlist.

• Let the question and findings determine the report structure.

Useful incomplete leads belong in the main findings with a concise uncertainty label and source, not hidden under coverage gaps. Reserve coverage gaps for genuinely unsearched or inaccessible areas. Lack of funding, founder identity, or complete product detail is not a reason to suppress a substantive signal.

• A name-only mention or ambiguous identity is an early signal when relevant:

give the observed name, why it matters, the source and what is unknown in one compact entry. Do not infer that an unverified project is fake or nonexistent. Low interaction, no third-party coverage, missing founder names or an unclear startup stage alone do not justify exclusion. Label promotional/self-reported material and any concrete contradictory evidence separately. Avoid unsupported claims like "the most obscure project on the internet" or market-wide cooling inferred from a single post. Do not print internal coverage approval/gate text.

• Build an opportunity map around problems, approaches, builders and emerging

signals. Include a compact candidate list with why each matters, an observed source link, uncertainties and a worthwhile next step. Scale detail to the request and budget; do not require a large report or graph for every answer.

• Connect people, projects and organizations using observed links. Distinguish

author, founder, employee, curator and guest. Preserve uncertain identity matches as hypotheses; do not merge on name alone. Show a relationship table or graph only when useful, with sources for asserted relationships.

• With multiple allowed platforms, follow promising links across them when

useful: a demo can lead to a repository, a builder profile or an interview. Do not verify every candidate on every platform. Reposts are not independent confirmation. A podcast description is not a listened-to transcript; a search preview is not a full read. Separate source claims, observed activity and inference, with relevant dates and contradictions.

• For tracking, suggest specific signals, sources and a sensible cadence. These

are recommendations, not a running monitor. For meeting preparation, suggest focused questions tied to unknowns; draft outreach only when requested and never send or contact anyone without explicit authorization.

• For explicit diligence, investigate the user's decisive questions and keep

missing evidence visible. Do not promote a promising lead to a verified, investable or fundraising company. Discovery does not require diligence first.

When update_research_board is available, record new leads as they arrive and show what is being followed on each platform. Use signal/following for incomplete findings; context for familiar comparables. In deep mode, publish a findings table (CSV or XLSX) covering both priority and early leads, with source links, uncertainties and next steps. Do not make table delivery a reason to prune leads.

Research lenses

Choose lenses that fit the user's question; these are optional angles, not a checklist or a fixed report structure. Let useful discoveries guide follow-ups.

• **Overlooked teams:** connect founders, products and community feedback across

sources to find less-exposed teams with meaningful early traction. Explain why a meeting could be worthwhile; distinguish usage from attention and self-reported growth. Compare with the user's known projects when available.

• **Founder portrait:** trace public work, interviews, posts and collaborations

over time. Look for technical preferences, changing views and contradictions, beyond a résumé. Cite da