@autonav/communication-layer

Protocol schemas and validation for navigator responses.

Install

``bash
npm install @autonav/communication-layer
`

What's in here

- Schemas - Zod schemas for responses, configs, queries
- Validation - Source checking, hallucination detection
- Errors - Typed errors for common failure cases
- Templates - CLAUDE.md templates and partials for navigator initialization
- Skills - Built-in skills (mend, commit, etc.) for navigator capabilities

This is protocol-only. No execution logic. For that, see @autonav/core.

Usage

`typescript
import {
NavigatorResponseSchema,
validateResponse,
createAnswerQuestionPrompt,
generateClaudeMd,
getAllSkills,
} from "@autonav/communication-layer";

// Validate a response
const result = validateResponse(response, knowledgeBasePath);
if (!result.valid) {
console.error(result.errors);
}

// Parse a response
const parsed = NavigatorResponseSchema.parse(json);

// Generate a CLAUDE.md template
const claudeMd = generateClaudeMd({
name: "my-navigator",
domain: "Platform Engineering",
knowledgeBase: "./knowledge",
});

// Get available skills
const skills = getAllSkills();
`

Key schemas

NavigatorResponse - What the navigator returns:
`typescript
{
query: string;
answer: string;
sources: Source[];
confidence: "high" | "medium" | "low";
outOfDomain: boolean;
}
`

NavigatorConfig - Navigator settings (config.json):
`typescript
{
version: string;
name: string;
knowledgeBase: string;
confidenceThreshold?: number;
}
`

Validation

validateResponse() checks:
- All cited source files exist
- No hallucination patterns (fake ARNs, placeholder paths)
- Confidence is reasonable

Returns { valid: boolean, errors: [], warnings: [] }`.