Skills

Skills are Allium’s internal instructions, exposed to your agent on demand. Instead of guessing how to structure a Snowflake query against a multi-billion-row chain table, your agent fetches the rules Allium’s own Assistant follows.

Call get_skill(name="sql-optimization") before writing or running any SQL. Allium’s tables are large enough that a query missing its partition filter can scan a full chain — the skill covers the pruning rules that prevent it.

Available tools

Tool	Description
`list_skills`	List the skills your organization can access, with descriptions
`get_skill`	Fetch a skill’s full instructions, or one of its reference files

Available skills

Skill	What it covers
`sql-optimization`	Partition pruning, default time ranges, CTE filtering, `QUALIFY` for dedup, `UNION ALL` over `UNION`, `APPROX_COUNT_DISTINCT`, pre-aggregated `.metrics.` tables, EVM address lowercasing, per-chain versus `crosschain.*` tables, Solana voting/non-voting views
`dashboard-design`	The `set_dashboard` build order, the element catalog, 12-column layout, and the design and SQL rules that make a dashboard read well
`explorer-visuals`	Per-type `spec` JSON schemas for every Explorer visual type
`product-guide`	Explorer, Realtime APIs, Datastreams, and Datashares — what each is for, and the canonical dated list of supported chains

For “what chains does Allium support?” questions, fetch product-guide and cite its list rather than answering from the model’s memory. The list is dated and maintained.

Fetching a skill

{
  "name": "get_skill",
  "arguments": { "name": "sql-optimization" }
}

Returns the full instructions as text. If the skill has reference files, they’re listed by title at the end of the body.

Reference files

Fetch one by passing its title exactly as listed:

{
  "name": "get_skill",
  "arguments": {
    "name": "explorer-visuals",
    "reference_title": "chart"
  }
}

This two-step shape keeps the main skill small. dashboard-design and explorer-visuals both use it: the skill body explains the approach, and each reference returns the exact JSON schema for one element or visual type.

Read the skill

get_skill(name="explorer-visuals") — general advice and the list of available types.

Read the type's reference

get_skill(name="explorer-visuals", reference_title="sankey") — the exact spec schema.

Build the spec

Construct spec against that schema and pass it to create_explorer_visual.

Discovering skills

{ "name": "list_skills", "arguments": {} }

Returns each accessible skill’s name and description. Individual skills can be gated, so this returns only what your organization can use — if a skill you expect is missing, contact [email protected].

Cache them locally

Skill content is stable per release. If your client supports persistent instructions, save a skill once instead of refetching it every session:

Client	Where
Claude Code	`.claude/skills/<name>/SKILL.md`, references under `.claude/skills/<name>/references/`
Other clients	Your client’s equivalent instruction or rules directory

Fetched skill bodies include this reminder, so an agent that reads one knows it can keep it.

SQL & queries

Where sql-optimization gets applied

Dashboards & sharing

Where dashboard-design and explorer-visuals get applied

Product comparison

The human-readable version of product-guide

Knowledge & discovery

Find the tables before you apply the rules

Assistant

MCP

Machine Payments

Available tools

Available skills

Fetching a skill

Reference files

Discovering skills

Cache them locally

SQL & queries

Dashboards & sharing

Product comparison

Knowledge & discovery

​Available tools

​Available skills

​Fetching a skill

​Reference files

​Discovering skills

​Cache them locally

​Related resources

SQL & queries

Dashboards & sharing

Product comparison

Knowledge & discovery

Available tools

Available skills

Fetching a skill

Reference files

Discovering skills

Cache them locally

Related resources