ARS 3.0 • Specification & ManualCLI v3.0.017 MCP Tools • 9 SkillsAPI v3.0 REST & JSON-RPC

Glintbase Documentation

The definitive technical guide to auditing, simulating, and hardening software for autonomous AI coding agents. Covers the 119-check ARS 3.0 standard, the multi-agent Flight Simulator, the 17-tool MCP server, and machine-first API endpoints.

01 / Command Interface

CLI Suite (`@glintbase/cli`)

npm @glintbase/cli

Install globally via npm i -g @glintbase/cli or execute instantly via npx @glintbase/cli. Requires Node.js ≥ 20.

glintbase audit [target] [options]

Core Auditor

Executes the 119-check ARS 3.0 standard across 4 layers with dynamic denominator scaling, archetype classification, and AST static analysis.

Available Options & Flags

Flag	Function
--simulate	Spawns embedded Flight Simulator alongside audit to empirically test mission completion
--report <file>	Exports complete executive Markdown audit report with remediation code diffs
--fail-under <score>	Exits with code 1 if ARS score drops below threshold (e.g. --fail-under 80)
--json	Outputs raw structured JSON telemetry for programmatic pipeline consumption

Usage Example

glintbase audit

bash

# Audit any public web API or documentation site
npx @glintbase/cli audit https://api.stripe.com

# Audit local codebase offline via AST inspection with strict threshold & simulation
glintbase audit . --simulate --fail-under 85 --report audit-report.md

02 / Empirical Verification

Autonomous Agent Flight Simulator

Instead of relying solely on static HTML regex checks, Glintbase spawns empirical multi-turn agent journeys. The simulator measures how real coding agents navigate your docs, understand your API parameters, and execute tools without burning token budgets.

claude-code200k Context

Deep schema traversal, multi-step dependency chaining, strict JSON Schema type checking. Evaluates whether nested parameter definitions cause tool-use stalls.

cursor128k Context

Fast semantic indexing, tight token budgeting, IDE code-completion emulator. Measures context burn and checks for concise inline examples.

perplexityZero-JS SSR

Zero-JavaScript SSR crawler, strict markdown parser, anti-SPA canary detector. Tests whether client-side hydration shells blind autonomous search agents.

The Token Tax Equation

Dollar Tax = (Prompt_tok × $3/1M) + (Completion_tok × $15/1M)

Unoptimized documentation loaded into agent context windows burns thousands of tokens per turn. At scale across a 50-engineer engineering team, documentation token tax averages $433,200 / year.

Schema Friction Index

Friction = (TypeMismatches × 0.4) + (MissingDescriptions × 0.3) + (AmbiguousNames × 0.3)

Measures the probability of tool execution failure before runtime. A friction index > 0.25 triggers agent hallucinations and infinite retry loops.

In-Chat Multi-Modal Experience & Visual Replay

When executed via MCP in Claude, Cursor, or Windsurf, the Flight Simulator delivers an interactive visual experience:

1. SVG Journey Tree

Rendered inline directly in the chat using MCP image blocks (image/svg+xml).

2. Compressed State URL

Zlib-deflated base64url hash (#data=...) linking to the full visual cockpit.

3. Claude Artifacts

Interactive SVG/HTML side-panels generated on-the-fly for inspection without leaving the conversation.

03 / Evaluation Science

Mathematical Scoring Engine (Dynamic Denominator)

Unlike legacy static audits that penalize sites for capabilities they don't need, ARS 3.0 uses a mathematically grounded Dynamic Denominator with archetype-based baselines and unpenalized bonus upside.

ARS 3.0 Governing FormulaStandard v3.0.0-spec

Final Score = min(100, round((S_base + S_bonus) / D_active × 100))

where Active Denominator: D_active = D_archetype_base + S_bonus_earned

Unpenalized Bonus Rule:

Bonus specifications (e.g. WorkOS auth.md, WebMCP, Google AP2) add 0 to the denominator when missing, preventing unfair score degradation.

Complementary Commerce Rule:

Supporting any verified protocol (x402, UCP, ACP) awards full payment cluster credit; unused payment protocols are marked N/A, never failed.

Archetype Base Denominators (D_archetype_base)

Archetype	Applicable Services	Eligible Layers	Base Denominator
api_devtool	APIs, SDKs, CLI tools, SaaS backend	Layer 1, 2, 3	85 pts
docs_kb	Documentation hubs, Knowledge bases	Layer 1, 2, 3 (Docs-kind)	85 pts
ecommerce	Commerce APIs, checkout surfaces	Layer 1, 2, 3, 4	100 pts
content_media	Publishers, corporate blogs, portals	Layer 1, 2	50 pts

04 / Protocol Standard

The 119-Check Taxonomy Across 4 Layers

Every check in ARS 3.0 evaluates a discrete failure surface that causes AI agents (Claude Code, Cursor, Perplexity) to stall, burn context tokens, or hallucinate.

Layer 1: Discovery & Entrypoints

Evaluates whether autonomous agents can find, index, and recognize your product surfaces across internet directories and AI crawlers.

Representative Protocol Checks in this Layer

robots.txt Agent Ingestion SignalsRequired

Explicit Allow policies for ClaudeBot, PerplexityBot, and GPTBot without deceptive wildcard Disallow.

8 pts max

Agent Resource Discovery (ard.json)Recommended

Valid /.well-known/ard.json manifest pointing to machine endpoints.

6 pts max

Canonical Root ResolutionRequired

Deterministic resolution of docs root without infinite redirects or domain aliasing.

6 pts max

Wikidata Entity & Authority LinkEmerging

Verified Wikidata Q-identifier providing authoritative grounding.

4 pts max

Full 119-check specification source catalog available in src/core/checks/registry.ts.

05 / Agent Tooling

Official MCP Server (17 Tools & 9 Skills)

Connect Glintbase to AI coding agents (Claude Code, Cursor, Windsurf, ChatGPT) to enable real-time readiness auditing, automated flight simulation, and self-healing code remediation inside your IDE.

1. Hosted Cloud MCP

Single endpoint with sliding-window rate limiting (60 req/min):

https://scan.glintbase.dev/api/mcp

2. Local Stdio Mode

Zero network overhead, direct stdin/stdout IPC for Claude Code & Cursor:

npx @glintbase/cli mcp

3. Quick Tunnel (--share)

Ephemeral secure Cloudflare Quick Tunnel for remote collaboration:

glintbase mcp --share

Complete 17-Tool Production Catalog

All tools expose typed JSON Schema parameters with strict Zod validation.

glintbase_audit50-120 tok

Runs full 119-check ARS 3.0 audit against any public URL or local repo.

Auditing• ARS 3.0 Scorecard JSON

glintbase_get_score10-25 tok

Retrieves cached ARS score, archetype baseline, and layer breakdown.

Auditing• Score summary

glintbase_discover_surfaces30-60 tok

Discovers canonical agent entrypoints (/llms.txt, /auth.md, ard.json, OpenAPI).

Auditing• Surfaces array

glintbase_simulate_flight250-450 tok

Executes multi-turn agent simulation with live Journey Tree SVG and replay hash.

Simulation• SVG + State Hash + Artifacts

glintbase_calculate_token_tax40-70 tok

Calculates prompt/completion token waste and annualized financial penalty.

Simulation• Dollar burn metric

glintbase_check_schema_friction35-65 tok

Analyzes OpenAPI/JSON Schema for missing types, vague descriptions, and ambiguity.

Simulation• Friction Index (0-1)

glintbase_counterfactual_proof120-200 tok

Proves score upside (+pts) from applying candidate AST remediations in-memory.

Simulation• Delta diff JSON

glintbase_generate_artifact80-150 tok

Generates production-grade living artifacts (llms.txt, auth.md, ard.json, canary).

Sandboxing• File content

glintbase_sandbox_validate60-110 tok

Validates documentation curl commands and SDK calls in isolated micro-sandbox.

Sandboxing• Execution result

glintbase_inspect_webmcp40-80 tok

Inspects client-side window.modelContext tools for visiting AI browser agents.

Sandboxing• WebMCP manifest

glintbase_ci_gate30-50 tok

Evaluates pull request against score threshold and computes PR drift delta.

Security• Pass/Fail boolean

glintbase_compliance_report90-180 tok

Generates executive compliance breakdown for enterprise security audits.

Security• Markdown report

glintbase_audit_canaries40-75 tok

Probes randomized paths to verify authentic HTTP 404 vs soft-200 SPA shells.

Security• Canary HTTP status

glintbase_verify_agent_auth35-60 tok

Validates WorkOS auth.md, RFC 9728 metadata, and token exchange flows.

Security• Auth diagnostic

glintbase_audit_mutation_safety45-80 tok

Verifies Idempotency-Key headers and safe mutation locks on write endpoints.

Security• Idempotency audit

glintbase_get_skill60-120 tok

Retrieves complete skill instruction playbook by URI (skill://glintbase/*).

Skills• Skill playbook MD

glintbase_install_skill20-40 tok

Scaffolds skill playbook into .agents/skills/ or ~/.claude/skills/ directory.

Skills• Install path

06 / Machine Integration

API Reference (REST & JSON-RPC 2.0)

Complete specification for Glintbase cloud services. AI agents and backend systems can query endpoints directly with standard JSON payloads. All machine endpoints return standard rate-limit and status headers.

POSThttps://scan.glintbase.dev/api/mcp60 requests / min sliding-window per IP

Model Context Protocol JSON-RPC 2.0 Gateway

Standard JSON-RPC 2.0 endpoint implementing official MCP protocol methods: initialize, tools/list, and tools/call. Dispatches to all 17 production tools with multi-modal SVG image and Claude artifact responses.

Request Parameters / Headers

Name	Type	Required	Description
jsonrpc	string	Yes	Must be exactly '2.0'
id	string \| number	Yes	Unique client request identifier
method	string	Yes	'initialize' \| 'tools/list' \| 'tools/call'
params	object	No	Method parameters (e.g. { name: 'glintbase_simulate_flight', arguments: { url: 'https://api.example.com' } })

Executable Code Examples

Model Context Protocol JSON-RPC 2.0 Gateway Example

curl

curl -X POST https://scan.glintbase.dev/api/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": "req-1",
    "method": "tools/call",
    "params": {
      "name": "glintbase_simulate_flight",
      "arguments": {
        "url": "https://docs.stripe.com",
        "agent": "claude-code",
        "intent": "Create a customer and attach payment method"
      }
    }
  }'

Expected Response Payload (HTTP 200)

Response JSON

json

{
  "jsonrpc": "2.0",
  "id": "req-1",
  "result": {
    "content": [
      {
        "type": "text",
        "text": "Flight Simulation complete for https://docs.stripe.com\nScore: 94/100 | Tax: $0.012"
      },
      {
        "type": "image",
        "data": "PHN2ZyB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHdpZHRoPSI4MDAiIGhlaWdodD0iNDAwIj4...",
        "mimeType": "image/svg+xml"
      }
    ],
    "structuredData": {
      "score": 94,
      "tokenTax": 0.0124,
      "frictionIndex": 0.08,
      "replayHash": "eJzLSM3JydErys9PLVbISUxPjU8rSgUAXYMHxw=="
    }
  }
}

07 / CI / CD Integration

GitHub Actions PR Drift Shield

Add Glintbase to your pull request pipeline. Automatically audits documentation and API modifications, tests PR drift against the base branch, and posts a Markdown scorecard comment directly onto the pull request.

.github/workflows/agent-readiness.yml

yaml

name: Agent Readiness Quality Gate

on:
  pull_request:
    branches: [main]
  push:
    branches: [main]

jobs:
  audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: 20

      - name: Run Glintbase ARS 3.0 Drift Shield
        run: |
          npx @glintbase/cli ci . \
            --fail-under 80 \
            --pr-drift \
            --github-token ${{ secrets.GITHUB_TOKEN }}

08 / Living Implementation