AI-Assisted Development

Scalus ships first-class support for LLM coding agents (Claude Code, Cursor, Codex, Copilot, and friends). Three layers work together: fetchable context artifacts on this site, AI-ready project templates, and a Claude Code plugin with task skills.

Context artifacts (llms.txt)

LLMs trained before Scalus 1.0 hallucinate outdated APIs. These plain-text artifacts give any agent the current ground truth:

URL	Content	When an agent should fetch it
/llms.txt	Index of everything below	First contact
/llms-api.txt	Version-pinned public API signatures: prelude, builtins, validator traits, ledger, TxBuilder, Emulator, testkit	Before writing any Scalus code
/llms-examples.txt	21 complete example validators with their tests	When writing a new validator or test
/llms-full.txt	All documentation pages as one markdown file	Deep dives
/llms-ts.txt	The npm package README and the full TypeScript definitions (`scalus.d.ts`)	Before writing TypeScript or JavaScript that uses the npm package `scalus`

Every documentation page is also available as plain markdown: append .md to its URL, e.g. /docs/smart-contracts/validators.md .

All artifacts are generated from the source of truth on every release – the API cheatsheet comes from the compiled code itself, so it cannot drift from the published library.

TypeScript and JavaScript projects

/llms-api.txt and /llms-examples.txt describe the Scala API. A TypeScript or JavaScript agent needs /llms-ts.txt instead. Once scalus is installed from npm, the same two files are in node_modules/scalus/ (scalus.d.ts and README.md) at the version the project uses. Tell the agent to read those first. scalus.d.ts is generated from the compiled code, like the Scala cheatsheet.

AI-ready project templates

Projects scaffolded from the hello.g8 or validator.g8 templates are AI-ready out of the box:

AGENTS.md – the cross-tool agent instruction file: build/test commands, the on-chain subset rules (@Compile restrictions, toData comparisons, prelude-only imports), common pitfalls, and the artifact URLs above. Claude Code, Cursor, and most agents read it automatically. CLAUDE.md points at it.
.claude/settings.json – registers the Scalus marketplace and enables the Scalus Claude Code plugin (below). Install the plugin once with /plugin install scalus@scalus.

Claude Code plugin

For existing projects, install the skills with the Scalus plugin:


/plugin marketplace add scalus3/scalus
/plugin install scalus@scalus

In a Scalus project, the plugin adds a routing table to each session at start-up, so the agent picks the right skill before it writes code. A Scalus project is a Scala build that names Scalus, or a package.json that depends on the npm package scalus. The six task skills load when the task matches:

Skill	Use
`scalus:contract`	Writing validators
`scalus:contract-test`	Testing validators
`scalus:local-development`	Emulator + TxBuilder development loop
`scalus:optimize-contract`	Execution-budget optimization review
`scalus:smart-contract-security-review`	Pre-deploy security audit
`scalus:typescript`	The npm package from TypeScript or JavaScript: Emulator, script evaluation, parameters

Skills are plain markdown in scalus-skills/ – agents other than Claude Code can read the SKILL.md files directly.

Tips for best results

Tell your agent to fetch /llms-api.txt before writing code and check every signature it plans to use. This is the single highest-impact instruction.
Point it at /llms-examples.txt and ask it to imitate the HTLC example – agents follow working code better than prose.
Have it measure execution budgets with the testkit assertions (assertBudgetWithin) instead of guessing.
Ask for negative tests: every validator test suite should prove the failure cases fail.