GP Research Assistant

An opinionated pre-meeting recipe for LPs. Give the skill a GP name and a time window, and it runs a trailing-window scan of firm, team, and portfolio-company news from public sources (GP site, Google News, PE trade press, LinkedIn, regulator filings), then renders a single-page HTML brief with six tabs, per-finding talking points framed as meeting questions, and LP-side next steps. Every finding is source-cited with URL, source type, dates, and a verbatim snippet. Portfolio news is filtered to an eight-item material-event taxonomy (exits, funding, C-suite, layoffs, pivots, regulatory, lawsuits, distress), so routine product launches and marketing chatter never make the brief.

What it does

• Trailing-window scan (default 6 months) across the GP site, Google News, PE trade press, public LinkedIn, and regulator filings
• Auto-discovers subjects: the firm itself, named partners from the team page, and current portfolio companies
• Filters portco news through an eight-item material-event taxonomy so routine launches and marketing never crowd the brief
• Sources every claim with URL, source type, publication date, retrieval date, and a verbatim snippet, no exceptions
• Runs a local-language pass automatically when the GP HQ is non-English (Nordic, German, French, Dutch, Italian, Spanish, and more)
• Generates one talking point per finding, framed as a meeting question with a named subject, plus one LP-side next step
• Outputs a single standalone HTML file with six tabs and per-finding drawers, ready to email or open offline

How to use

1. 1

   Download the skill

   Click Download. The .skill file is a ZIP bundle containing a single SKILL.md, the workflow itself in plain Markdown.

2. 2

   Install it in your AI of choice

   In Claude, open Settings, Customize, Skills, and upload the .skill file directly. For Copilot or other LLMs, rename it to .zip, extract SKILL.md, and paste its contents into the agent's instructions or system prompt.

3. 3

   Start the conversation

   Once installed, just ask. The skill's scoping questions kick in automatically and the model follows the recipe end to end.

Limitations

• Public sources only. Does not hit LexisNexis, World-Check, Dow Jones, Sayari, court dockets, or OpenCorporates
• Not diligence and not KYC. Every brief carries a disclaimer to that effect, and material findings should be verified before acting
• No persistence in this version. Each run is a fresh snapshot, not a diff against the last brief
• Outside FundFrame, the skill runs ad-hoc against whatever GP name you paste in. Inside FundFrame, it runs natively in the Investment CRM against your live GP records, pulls the meeting context from your calendar, and stores the brief on the GP's timeline so the next LP on the team picks up where you left off

Changelog

• v0.2.1June 2026Talking points enforced as meeting questions, not topics. Next steps tightened to concrete LP-side actions.
• v0.2.0May 2026Six-tab HTML brief with per-finding drawers, eight-item material-event taxonomy for portco news, auto local-language pass.
• v0.1.0May 2026Initial release: trailing-window public-news scan on a single GP, single-page HTML output.