Skip to content

hsaidno-cpu/skill-analyzer

v1.2.0

Sistem analisis modular dengan Universal AI Invocation Gateway, Remote MCP, HTTP API, autentikasi OIDC, routing terpusat, dan logging terkontrol.

GitHub

SKILL ANALYZER 1.2.0

A modular, evidence-led analysis system with a portable Agent Skills core and a Universal AI Invocation Gateway.

SKILL ANALYZER combines ten analytical specialists, five shared engines, and two utility skills. The Universal AI Invocation Gateway adds a canonical command registry, platform invocation adapters, a JSON envelope contract, and a local prompt compiler.

What changed in 1.2.0

  • Adds a deployable cloud runtime for Remote MCP and JSON HTTP API on one public origin.
  • Adds OAuth 2.1/OIDC JWT verification, centralized command routing, OpenAI-compatible model execution, Redis rate limiting, and metadata-only audit logging.
  • Adds a Docker image, cloud deployment guide, remote-gateway profile, and cloud integration tests.
  • Local adapter installation is optional when using the hosted service: callers connect to the remote MCP endpoint or HTTP API.

What changed in 1.1.0

  • Adds the Universal AI Invocation Gateway with eleven canonical commands and alias normalization.
  • Adds platform profiles for Codex, Claude Code, GitHub Copilot surfaces, Gemini CLI, Cursor, VS Code, OpenAI API, ChatGPT package, and generic Agent Skills clients.
  • Adds a versioned JSON invocation envelope and schema.
  • Adds a dependency-free CLI to resolve commands and generate platform-ready prompts or JSON.
  • Adds a Gemini CLI extension manifest and eleven custom command shortcuts.
  • Extends the exporter to ship Gateway contracts and invocation tooling with each profile.
  • Extends static validation and CI with routing, alias, adapter, and command tests.

Core architecture

The master orchestrator routes to ten analytical responsibilities: analytical-collector, analyzer, negative-collector, positive-converter, evidence-auditor, strategic-foresight, evaluator, komparator, comparative-analytic, and information-authenticator. Five engines provide shared procedures for decisions, human impact/governance, benchmark/regression, knowledge/method intelligence, and FIRE.

The two utility skills, platform-compatibility and invocation-gateway, maintain distribution and access. They are not additional analytical specialists.

Quick start

# Validate the package and Gateway contracts
python scripts/validate_portability.py

# List available commands
python scripts/invoke.py list

# Generate a prompt for any supported host profile
python scripts/invoke.py resolve audit --target "Your system or document" --platform generic --depth deep

# Generate machine-readable invocation JSON
python scripts/invoke.py resolve compare --target "Option A versus Option B" --platform openai-api --format json

# Export platform-specific skills and Gateway adapters
python scripts/export_platforms.py --target all --output ./dist

# Run routing contract tests
python -m unittest discover -s tests -v

See GATEWAY.md for command syntax, installation, platform-specific invocation, and the JSON contract. See PORTABILITY.md for distribution profiles and compatibility boundaries.

Gemini CLI quick install

The repository root includes a Gemini extension manifest, context entrypoint, bundled Agent Skills, and custom prompt commands:

gemini extensions install https://github.com/hsaidno-cpu/skill-analyzer

After installation, use commands such as /analyzer:audit, /analyzer:compare, or /analyzer:forecast. Verify the extension and skills with the relevant Gemini CLI list commands.

Reliability boundary

The Gateway normalizes intent and prepares routing. The CLI does not call an AI model, grant platform tools, spawn agents, or persist execution traces. A passing package or static CI job does not prove that a particular host discovers a skill, selects the correct model behavior, executes a subagent, or produces correct conclusions. Platform-specific behavior must be tested and recorded against the exact host, model/provider, version, product surface, and evidence.

Hosted Universal Cloud Gateway

The repository includes a deployable service package:

  • Remote MCP (Streamable HTTP): https://YOUR_DOMAIN/mcp
  • HTTP API: https://YOUR_DOMAIN/v1/invoke
  • Command discovery: https://YOUR_DOMAIN/v1/commands
  • Health/readiness: /healthz and /readyz

Both protocols share one public base URL. A caller that uses the hosted service does not need to install local SKILL ANALYZER adapters or copy the skill directory. It needs an OAuth 2.1/OIDC access token with the analyzer:invoke scope.

See CLOUD_DEPLOYMENT.md for local setup, production secrets, OIDC configuration, Redis, and a Cloud Run reference deployment. See CLOUD_ARCHITECTURE.md for trust boundaries.

Deployment status: the service implementation is committed, but this repository has not provisioned a public cloud endpoint. A cloud account, domain, OIDC issuer, model API key, Redis service, and deployment operation are required before a public URL exists.