@rsconcept/rstool

Agent-facing library for incremental RSForm construction, RSLang expression analysis, diagnostics, modeling, and evaluation. It wraps @rsconcept/domain with a deterministic session contract and a stdio JSON wrapper that LLM agents and MCP clients can call directly.

Agent skill

RS language + rstool workflows for agents: skills/rstool-helper/ (GUIDE.md, REFERENCE.md, EXAMPLES.md) and docs/ (DOMAIN.md, CONCEPTUAL-SCHEMA.md, SYNTAX.md, …). A thin entry skill is installed into the project’s agent skills folder; see Installing the agent skill below.

Install

npm install @rsconcept/rstool

@rsconcept/domain is a peer-of-dependency installed automatically.

Installing the agent skill

After npm install @rsconcept/rstool, you do not copy files yourself. Ask your agent, for example:

Install the rstool agent skill according to the package instructions.

The agent should read and follow:

node_modules/@rsconcept/rstool/skills/INSTALL.md

That copies skills/rstool-helper/SKILL.md from the package into your host’s project skills directory (for example .agents/skills/rstool-helper/SKILL.md). Full guidance stays in the package (skills/rstool-helper/GUIDE.md, docs/*.md) and is read from node_modules when needed.

Details: skills/README.md.

Quick use (library)

import { CstType, RSToolAgent } from '@rsconcept/rstool';

const tool = new RSToolAgent();
const session = tool.createSession();
const result = tool.analyzeExpression(session.sessionId, {
  expression: '1+2',
  cstType: CstType.TERM
});

Quick use (stdio wrapper)

npx rstool-wrapper      # one JSON request per line on stdin, one JSON response per line on stdout

Or run it as a child process from your own code:

import { RSToolWrapperClient } from '@rsconcept/rstool/wrapper';

const client = new RSToolWrapperClient(); // spawns `rstool-wrapper` by default
await client.waitUntilReady();
const session = await client.call<{ sessionId: string }>('createSession');
await client.close();

Scope

• Session-based incremental editing of constituents.
• Parse / syntax / semantic / type analysis for expressions.
• In-memory modeling: set base bindings and structured values; evaluate expressions and constituents.
• Deterministic diagnostics and export/import for reproducible agent workflows.

Publishing

Maintainers: see PUBLISHING.md for npm release steps.

Repo scripts

From rsconcept/rstool (or run powershell -File scripts/dev/LocalDevSetup.ps1 from the repo root on Windows):

• npm install / npm ci
• npm run typecheck
• npm test
• npm run build — produce dist/ via tsdown
• npm run wrapper — dev stdio wrapper via tsx
• npm run example:client, npm run example:build-schema, npm run example:build-rsmodel — полный список example:* и kinship:cli в examples/README.md

Stdio protocol

• Input: one JSON request per line.
• Output: one JSON response per line.
• The wrapper keeps state in memory while the process is alive.
• On startup, a ready handshake is printed.

Supported methods (current contract version: see CONTRACT_VERSION):

• ping
• methods
• createSession
• addOrUpdateConstituenta
• analyzeExpression
• getFormState
• listDiagnostics
• commitStep
• exportSession
• exportPortalSchema
• exportPortalModel
• importSession
• setConstituentaValue
• setConstituentaValues
• clearConstituentaValues
• getModelState
• evaluateExpression
• evaluateConstituenta
• recalculateModel

Example request:

{ "id": "1", "method": "createSession", "params": {} }

Example response:

{ "id": "1", "ok": true, "result": { "sessionId": "...", "contractVersion": "1.4.0" } }

Typed client example

Run:

npm run example:client

File: examples/agent-client.ts

The example:

• starts the stdio wrapper as a child process
• waits for the ready handshake
• creates a session
• upserts a constituent
• runs expression analysis
• sets a base binding and evaluates a term
• fetches diagnostics

License

MIT