DOCSHOW TO MODIFY PROMPTS TO TAILOR YOUR DASHBOARD LOCALLY

How to Modify Prompts to Tailor Your Dashboard locally

Loading last updated info...

SSW Tiger comes with sensible defaults, but every team is different. This guide shows you how to change what the dashboard looks like, what it analyses, and the tone it uses without writing any code.

Who is this guide for?

This guide is for users running SSW Tiger locally on their own machine. The edits below change what your local copy produces.

The prompt is just a few text files written in plain English. Edit them, run against a transcript, then check the result. That's the whole loop.

The Easiest Way: Just Ask Claude

You don't need to hunt through files or write any code. Open Claude Code in the SSW.Tiger folder and describe the change in plain English. Claude already knows where the prompt lives and can edit it for you.

Then say things like:

  • "Make the dashboard tone a bit softer. Use fewer hard truths and focus more on areas to improve."
  • "Remove the Trends tab. We only have one meeting so far."
  • "Add a Customer Quotes section to the Overview tab. Max 3 bullets."
  • "Replace SSW branding with Acme Corp. Our brand red is #1A73E8."
  • "Anonymise all participant names. Use Person 1, Person 2, etc."

Claude will figure out which files to edit (CLAUDE.md, the agent files, or the template) and make the changes. Then run a sample transcript again to see the result.

Customisation Ideas

Here are the most common things teams change. Try any of these as a Claude prompt, and rephrase them to match what you need.

Tone & Style

  • "Make the analysis less harsh and more constructive."
  • "Be sharper. Name names, quote the exact sentence, and avoid hedging language."
  • "Use British English everywhere."

Branding

  • "Rebrand the dashboard for [Your Company]. Primary colour is [#hex]. Replace the SSW logo motif."
  • "Use our team's profile photos from [URL pattern] instead of SSW People."

Structure

  • "Rename the Trends tab to History."
  • "Remove the Analytics tab. We don't need cost calculations."
  • "Add a Risks section to the Overview tab. Highlight anything blocking the next sprint."

Content Length

  • "Shorten everything. Max 3 bullets per section, max 1 sentence per bullet."
  • "Make insights longer. Give more context and examples for each one."

Privacy

  • "Anonymise participant names. Keep the same mapping across all tabs."
  • "Don't show client names anywhere. Replace them with Client A, Client B, etc."

Analysis Focus

  • "Stop calling out elephants in the room. Keep insights neutral."
  • "Add a metric for time spent off topic. Track how many minutes were spent on tangents."
  • "Track question response time. Show how quickly each person answered when asked something."

Tips for Better Results

  • Be specific. "Be more concise" is vague. "Max 2 sentences per insight, no adjectives" works.
  • Show, don't just tell. If you want a specific format, paste an example into the chat.
  • Change one thing at a time. If the output gets weird after five changes, you won't know which one caused it.
  • Test after each change. Run a transcript you know well, so you can spot what shifted.

What NOT to Change

Even if you ask Claude, it should leave these alone. They are important parts of the system:

  • The Consolidation step: this stops Tab 1 saying "Alice" while Tab 3 says "Product Owner"
  • The Participant Resolution rules: they prevent Claude from inventing people who weren't in the meeting
  • The Tailwind and Chart.js CDN links in the template: without them, styling breaks

If a change touches any of these, Claude will (and should) push back. Trust that.

Going Deeper

If you'd rather peek under the hood, the customisation lives in three files:

CLAUDE.md

Controls the dashboard structure, including tabs, sections, and rules.

.claude/agents/*.md

Controls how each AI analyst thinks, including tone, depth, and what to look for.

templates/dashboard.html

Controls the visual look, including fonts, colours, and layout.

You can open these directly and edit by hand. They're plain English, not code. For most teams, asking Claude is faster and safer.

Testing Your Changes

The cycle is fast and usually takes under a minute per try.

  1. Make one change.
  2. Ask Claude Code: "Process the test project transcript again with the new instructions."
  3. Open the dashboard and compare it with your sample meeting.

Pro Tip: Keep one sample .vtt you know really well. You'll spot what changed faster.

Additional Tips

  • Start by tweaking tone before changing structure. Most teams just want the analysis to "sound right"
  • Save a copy of CLAUDE.md before big edits, so you can always roll back
  • Read the existing rules first. CLAUDE.md already has rules like "no markdown files, only HTML" and "use canonical names". Conflicts cause messy output.