Skills

Bring your own model

Run a skill on any modern LLM

FundFrame skills are plain Markdown wrapped in a ZIP, so any model that accepts a system prompt or custom instructions can follow them. Here's how to use one with Gemini, Mistral, Perplexity, Grok, and others.

Before you start

What you need

  • An account with the LLM of your choice. Gemini, Mistral Le Chat, Perplexity, Grok, DeepSeek, and Llama-based assistants all work.
  • Access to either a custom system prompt, a Gem or Custom GPT-style template, or a long-form initial message, depending on the provider.
  • The skill file downloaded from the FundFrame skill page. The .skill file is a ZIP bundle, so rename the extension to .zip and extract it. The SKILL.md inside is what you paste into the model.

The walkthrough

Step by step

  1. 1

    Extract the skill bundle

    Rename the downloaded file from .skill to .zip, then double-click to extract. Inside the folder you'll find a single SKILL.md file. That markdown is the workflow itself, plain text, no transformation needed.

  2. 2

    Locate the system prompt field

    Every modern LLM exposes a way to seed the conversation. In Gemini it's the Gem instructions, with SKILL.md uploadable as a knowledge file too. In Mistral it's the Le Chat agent prompt. In Perplexity it's the Space description. In Grok it's the custom instructions field. In any model with an API, it's the system role message.

  3. 3

    Paste SKILL.md into the provider

    Open SKILL.md in any text editor, copy the full contents, and paste them as the system prompt or custom instruction. In Gemini Gems you can additionally upload SKILL.md under Knowledge so the Gem keeps the file alongside its instructions.

  4. 4

    Start the conversation

    Ask the model to begin. For Market Mapping: Build a coverage trip map for Copenhagen buyout funds. The skill's scoping questions should kick in automatically.

  5. 5

    Save the configuration

    Most providers let you name and persist the configured assistant, Gems, Spaces, Agents, Custom Models. Save it so colleagues can reuse the same skill without re-pasting.

Worth knowing

Tips

  • Models with smaller context windows may truncate long skills. If output quality drops mid-run, ask the model to recall the scoping section and continue.
  • Open-source models, Llama, Mistral, DeepSeek, work well when run on inference platforms that respect long system prompts. Local model UIs often don't, so check before importing.
  • Some providers, Perplexity Pro for example, switch models per turn. Pin a strong reasoning model for skill runs to get consistent output.

If something feels off

Troubleshooting

The .skill file will not open in a text editor
It is a ZIP archive. Rename the extension from .skill to .zip, then double-click to extract. The folder inside contains the SKILL.md you need.
The model ignores the skill and answers casually
Confirm you pasted SKILL.md into the system prompt or Gem instructions, not the first user message. If the provider doesn't expose a system prompt, prepend the contents to your first message and ask the model to follow it exactly.
Output is shorter than Claude or Copilot
Different providers have different default output limits. Ask the model to continue, or split the run across multiple turns. The skill is built to resume.
Outputs differ between providers
Expected. The skill specifies the workflow, but individual models bring their own tone, formatting habits, and knowledge cutoff. The structure should match, even if the prose differs.
Back to all skills