Skip to main content
Researches a market or technology space and renders an interactive, color-coded comparison matrix in the browser by combining live web research with a zero-dependency HTML table. Every cell is backed by a web search result — training data alone is never used for ratings. Use this skill when the user asks for competitive analysis, a competitive matrix, feature comparisons, market landscape, alternatives to X, or “who are the players in this space.”

​
Invocation

/competitive-matrix [market, product category, or technology space]
Examples: 
/competitive-matrix vector databases
/competitive-matrix open-source API gateways
/competitive-matrix LLM observability tools

​
Output

The skill writes a self-contained HTML file to /tmp/matrix.html and opens it in the browser. The matrix features:
  • Color-coded cells (green / blue / orange / red)
  • Sort on every column (click any header)
  • Filter by competitor name (case-insensitive)
  • Dark background (#1e1e2e)
  • A “Last updated: YYYY-MM-DD” line above the table

​
Rating scale

All ratings use the four-point scale exclusively. No numbers, percentages, stars, or prose in cells:
SymbolMeaningCell color
checkcheckExcellentGreen (#27ae60)
checkGoodBlue (#2980b9)
tildePartial / mixedOrange (#d35400)
xPoor / missingRed (#c0392b)
Every row ends with a Verdict column — a 5–10 word phrase summarising that player’s position.

​
Workflow

1

Identify the space

Confirm the market, product category, or technology being compared. Ask for clarification if the scope is ambiguous.
2

List the players

Enumerate 5–10 key competitors or options to include. If the user lists more than 10, select the 10 most relevant and note the exclusions — matrices with more than 10 rows become unreadable.
3

Choose evaluation dimensions

Select 5–8 columns relevant to this specific domain. Avoid generic filler columns. Dimensions should reflect what actually differentiates the players in this space.
4

Research each player

Run at least one web search per player to verify current status and ratings. Training data goes stale; search results do not. Never assign a rating without a supporting search result from this session.
5

Assign ratings

Score every cell using the four-point scale with evidence from the research step. Every cell must have a rating — no blanks.
6

Write the HTML file

Write a complete self-contained file to /tmp/matrix.html using the template from TEMPLATE.md. The file must include: color-coded cells, sortable columns, a filter input, the dark background, and the “Last updated” date.
7

Open in browser and screenshot

Call mcp__chrome-devtools__new_page with url: file:///tmp/matrix.html. Wait for the first competitor name to appear (wait_for, timeout 8000ms). Then take a screenshot.
Always use wait_for before taking the screenshot — not immediately after new_page. The page may not be fully rendered yet.
8

Deliver summary

After the screenshot, output 3–5 bullet points naming the key differentiators and standout winner(s) per use case.

​
Self-review checklist

Before delivering, verify all of the following:
  • At least one web search run per player — no player rated from training data alone
  • Between 5 and 10 players in the matrix
  • Between 5 and 8 evaluation dimension columns, plus a Verdict column as the last column
  • Every cell has a rating from the four-point scale — no blanks, no numbers
  • Every cell is color-coded with the correct color for its rating
  • A “Last updated: YYYY-MM-DD” line appears above the table
  • Sort works on every column
  • Filter input present and filters by competitor name (case-insensitive)
  • Dark background (#1e1e2e) applied
  • Screenshot taken after wait_for — not immediately after new_page

​
Golden rules

Run at least one web search per player before assigning any rating. Training data goes stale; search results do not.
The only valid ratings are checkcheck, check, tilde, and x. Never use numbers, percentages, stars, or prose in cells.
Every row must end with a 5–10 word phrase summarising that player’s position. The Verdict column is the last column.
A matrix where all cells look the same is broken. Color-coding is what makes the matrix readable at a glance.
Matrices with more than 10 rows become unreadable. If the user lists more, select the 10 most relevant and note the exclusions.
Competitive landscapes change. The date the research was conducted must appear above the table in the rendered output.

​
Reference files

FileContents
TEMPLATE.mdComplete HTML template with color-coded cell rendering, sort/filter JS, CSS design tokens, and rating scale constants
RESEARCH.mdHow to choose evaluation dimensions, research each player, assign ratings, and handle stale or conflicting data