Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when the user wants to persist a preference, skill, or knowledge. Use when it would aid future val development to store a memory of how best to build something.
.claude/skills/hashgraph-online-create-skill/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-11 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -17% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -36% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -23% | 0% |
Skills in Val Town instruct Townie and other AI agents using the Val Town MCP server how to write idiomatic vals that respect a user's preferences and knowledge.
In any val, a user can create a /skills/<name>/SKILL.md file, e.g. /skills/design/SKILL.md. Townie and the Val Town MCP server index skills with that directory/file structure across all of a user's vals. A user may choose to centralize their skills in one val or co-locate skills across multiple vals.
A skill markdown file must have frontmatter:
name: kebab-case name of the skill; contains lowercase letters, numbers, and hyphensdescription: helps the agent decide when the skill is relevant. A good skill description is critical for agent discoverytriggers: (optional) is a list of keywords to tip off the agentThe description and triggers fields enable skill discovery, i.e. tells the user's AI agent (e.g. Claude Code, Codex, Cursor) when to use it. The more specific the better, including key terms that should trigger use (which can also be enumerated in triggers). Skills without frontmatter will be silently skipped, so Townie/MCP will not be able to access them.
The Claude Platform Docs offer skill authoring best practices, including:
md--- name: design description: Use when styling a val's UI. Use for frontend vals that return JSX or HTML triggers: [css, styling, layout, theme] --- - Use `.css` files, avoid inline styles and Tailwind - Locate React components in a `/components` directory, one component per file - Use a sans-serif web-native font stack, no external fonts - ...
A new skill can be created by remixing another user’s skill and customizing the SKILL.md file. There is a remix button in the val.town UI, and a remix_val tool in the Val Town MCP server to do so.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-16 | pass→pass | 16,632 | 7,247 | -56% | 1 | 1 | 0% | 1,650 | 831 | -50% | 0 | 0 | — |
case-11 | fail→pass | 12,901 | 2,404 | -81% | 1 | 1 | 0% | 1,375 | 812 | -41% | 0 | 0 | — |
case-06 | fail→pass | 7,716 | 3,418 | -56% | 1 | 1 | 0% | 1,300 | 1,080 | -17% | 0 | 0 | — |
case-01 | fail→pass | 12,857 | 11,469 | -11% | 1 | 1 | 0% | 2,411 | 1,749 | -27% | 0 | 0 | — |
case-02 | fail→pass | 22,840 | 7,737 | -66% | 1 | 1 | 0% | 2,910 | 1,862 | -36% | 0 | 0 | — |
case-03 | fail→pass | 20,152 | 7,856 | -61% | 1 | 1 | 0% | 2,468 | 1,894 | -23% | 0 | 0 | — |
case-04 | fail→pass | 15,518 | 7,079 | -54% | 1 | 1 | 0% | 1,770 | 989 | -44% | 0 | 0 | — |
case-05 | fail→pass | 16,051 | 3,297 | -79% | 1 | 1 | 0% | 1,582 | 1,063 | -33% | 0 | 0 | — |
case-07 | pass→pass | 17,336 | 4,787 | -72% | 1 | 1 | 0% | 1,772 | 1,273 | -28% | 0 | 0 | — |
case-08 | pass→pass | 11,502 | 10,003 | -13% | 1 | 1 | 0% | 1,814 | 1,505 | -17% | 0 | 0 | — |
case-09 | fail→pass | 16,731 | 9,928 | -41% | 1 | 1 | 0% | 1,823 | 1,310 | -28% | 0 | 0 | — |
case-10 | pass→pass | 10,958 | 9,781 | -11% | 1 | 1 | 0% | 1,543 | 1,232 | -20% | 0 | 0 | — |
case-12 | fail→pass | 16,984 | 4,275 | -75% | 1 | 1 | 0% | 1,580 | 1,257 | -20% | 0 | 0 | — |
case-13 | fail→pass | 16,676 | 8,614 | -48% | 1 | 1 | 0% | 1,799 | 1,097 | -39% | 0 | 0 | — |
case-14 | pass→pass | 13,382 | 2,380 | -82% | 1 | 1 | 0% | 989 | 886 | -10% | 0 | 0 | — |
case-15 | pass→pass | 19,241 | 7,754 | -60% | 1 | 1 | 0% | 2,131 | 1,705 | -20% | 0 | 0 | — |
case-17 | fail→pass | 10,413 | 8,927 | -14% | 1 | 1 | 0% | 1,664 | 1,097 | -34% | 0 | 0 | — |
case-18 | pass→pass | 12,324 | 10,766 | -13% | 1 | 1 | 0% | 1,095 | 981 | -10% | 0 | 0 | — |
case-19 | pass→pass | 12,070 | 9,139 | -24% | 1 | 1 | 0% | 1,177 | 1,096 | -7% | 0 | 0 | — |
case-20 | pass→pass | 16,986 | 4,592 | -73% | 1 | 1 | 0% | 1,642 | 1,426 | -13% | 0 | 0 | — |
case-21 | pass→pass | 15,798 | 9,650 | -39% | 1 | 1 | 0% | 1,975 | 2,313 | +17% | 0 | 0 | — |
case-22 | pass→pass | 13,450 | 11,428 | -15% | 1 | 1 | 0% | 1,474 | 1,621 | +10% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +50 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.