ISL Discovery Engine
Automatic implementation discovery and binding generation for ISL (Intent Specification Language).
Overview
The ISL Discovery Engine automatically maps ISL specifications (behaviors, entities, types) to implementation code (functions, routes, classes) using multiple discovery strategies:
- Filesystem heuristics - Matches symbols based on file paths and directory structure
- AST scanning - Analyzes code structure to find exports, handlers, and routes
- Naming conventions - Matches symbols using naming pattern variations (camelCase, snake_case, etc.)
- Route matching - Maps behaviors to HTTP routes (e.g.,
Login → POST /api/login)
Features
- ✅ Automatic symbol extraction from ISL specs
- ✅ Codebase scanning for functions, routes, and exports
- ✅ Multi-strategy matching with confidence scores
- ✅ Bindings file generation (
.shipgate.bindings.json)
- ✅ Evidence tracking for each binding
Usage
CLI Command
isl bind specs/auth.isl
isl bind specs/auth.isl --impl ./src --min-confidence 0.5
isl bind specs/auth.isl specs/users.isl
isl bind specs/auth.isl --output .bindings.json
Programmatic API
import { discover, writeBindingsFile } from '@isl-lang/isl-discovery';
const result = await discover({
rootDir: './src',
specFiles: ['specs/auth.isl'],
minConfidence: 0.3,
verbose: true,
});
await writeBindingsFile('.shipgate.bindings.json', result, ['specs/auth.isl']);
Bindings Format
The discovery engine generates .shipgate.bindings.json files with the following structure:
{
"version": "1.0.0",
"timestamp": "2026-02-09T12:00:00.000Z",
"specs": ["specs/auth.isl"],
"bindings": [
{
"isl": {
"type": "behavior",
"name": "Login",
"domain": "UserAuthentication",
"specFile": "specs/auth.isl",
"location": {
"start": { "line": 30, "column": 1 },
"end": { "line": 74, "column": 1 }
}
},
"code": {
"type": "route",
"name": "POST /api/login",
"file": "src/routes/auth.ts",
"location": {
"start": { "line": 45, "column": 1 },
"end": { "line": 45, "column": 1 }
},
"metadata": {
"method": "POST",
"path": "/api/login"
}
},
"confidence": 0.85,
"evidence": [
{
"type": "route_matching",
"description": "Route path contains behavior name: POST /api/login",
"confidence": 0.80
}
],
"strategy": "route_matching"
}
]
}
Discovery Strategies
1. Filesystem Heuristics
Matches symbols based on file paths:
Login behavior → routes/login.ts or handlers/login.ts
User entity → models/user.ts or entities/user.ts
Confidence: 0.60-0.75
2. AST Scanning
Analyzes code structure to find:
- Fastify route registrations (
app.get(), app.post(), etc.)
- Exported functions (
export function, export const)
- Class exports (
export class)
Confidence: 0.70-0.85
3. Naming Conventions
Matches using naming variations:
CreateUser → createUser, create_user, CreateUser
- Handles camelCase, PascalCase, snake_case conversions
Confidence: 0.65-0.85
4. Route Matching
Maps behaviors to HTTP routes:
CreateUser → POST /api/users
GetUser → GET /api/users/:id
DeleteUser → DELETE /api/users/:id
Confidence: 0.80-0.85
Confidence Scores
Confidence scores range from 0.0 to 1.0:
- 0.9-1.0: Very high confidence (exact matches)
- 0.8-0.9: High confidence (strong patterns)
- 0.6-0.8: Medium confidence (reasonable matches)
- 0.3-0.6: Low confidence (weak patterns)
- <0.3: Filtered out by default
Options
rootDir - Root directory to search (default: current directory)
specFiles - ISL spec file(s) to bind
codeDirs - Code directories to search (default: all)
includePatterns - File patterns to include (default: **/*.ts, **/*.js)
excludePatterns - File patterns to exclude (default: **/node_modules/**, **/dist/**)
minConfidence - Minimum confidence threshold (default: 0.3)
enableAST - Enable AST scanning (default: true)
enableFilesystem - Enable filesystem heuristics (default: true)
enableNaming - Enable naming convention matching (default: true)
verbose - Verbose output (default: false)
Examples
Fastify Project
isl bind specs/auth.isl --impl ./src --code-dirs src/routes,src/handlers
Multiple Specs
isl bind specs/auth.isl specs/users.isl specs/payments.isl
Integration
The bindings file can be used by:
- Verification tools - To verify implementations against specs
- Code generation - To generate type-safe wrappers
- Documentation - To generate API documentation
- Testing - To generate test cases from specs
Limitations
- Requires consistent naming conventions for best results
- May produce false positives for generic names
- Confidence scores are heuristic-based, not guaranteed
- Complex codebases may require manual binding adjustments
Future Improvements