Skip to content

Repository files navigation

Cloud Decision Kit

English · 한국어

Validate License: MIT Python 3.11+

Evidence-first cloud architecture and TCO decision support for ChatGPT and Codex.

Cloud Decision Kit turns a workload into a small, defensible shortlist of deployable cloud architectures. It compares current official pricing and policies alongside operational burden, reliability, constraints, and lock-in—without requiring cloud credentials, provider SDKs, an MCP server, or a bundled price snapshot.

Why Cloud Decision Kit?

Cloud decisions are rarely settled by a single advertised unit price. The cheapest-looking service can become the wrong choice after egress, backups, observability, support, idle capacity, regional availability, quotas, or operator time are included.

Cloud Decision Kit helps you:

  • compare complete architectures instead of isolated product names;
  • eliminate candidates that fail hard requirements before scoring price;
  • model Low, Base, and Peak recurring cost plus 12-month TCO;
  • verify material claims against current, dated, official sources;
  • account for SLA, quotas, backup, residency, support, and exit risk;
  • identify the threshold that would reverse the recommendation.

The bundled source registry currently covers 24 providers and 123 official pricing or policy pages. Prices are intentionally not hard-coded; material facts are rechecked at decision time.

Supported workloads

  • static, edge, and content-heavy applications;
  • conventional web APIs and SaaS products;
  • mobile backends, PaaS, and BaaS;
  • relational, document, cache, and storage services;
  • event-driven and serverless workloads;
  • realtime collaboration and messaging;
  • authoritative multiplayer game backends;
  • data, batch, analytics, and warehouse workloads;
  • AI APIs, inference, training, and GPU workloads;
  • self-hosted and hybrid control planes.

How it works

  1. Build a workload card
    Capture traffic, regions, data, latency, availability, recovery, team, budget, and growth assumptions.

  2. Apply hard gates
    Remove options that fail non-negotiable region, protocol, quota, compliance, recovery, or operational requirements.

  3. Shortlist deployable architectures
    Compare complete combinations such as edge + serverless + managed database, rather than comparing product labels in isolation.

  4. Verify material evidence
    Recheck current official pricing, limits, regions, SLA terms, backup behavior, and policies that can change the decision.

  5. Normalize TCO
    Include compute, data, egress, backups, observability, support, idle capacity, and required add-ons across Low, Base, and Peak scenarios.

  6. Deliver a decision
    Recommend one architecture and one runner-up, with risks, confidence, unknowns, and reversal thresholds.

Installation

Recommended: Codex plugin marketplace

Register this repository as a marketplace source:

codex plugin marketplace add dd3ok/cloud-decision-kit
codex plugin marketplace list

For a reproducible install, pin a release tag:

codex plugin marketplace add dd3ok/cloud-decision-kit --ref v1.0.0

Refresh or restart the ChatGPT desktop app, open Plugins, select the Cloud Decision Kit marketplace, and install Cloud Decision Kit.

Test a local clone

git clone https://github.com/dd3ok/cloud-decision-kit.git
cd cloud-decision-kit
codex plugin marketplace add .

Direct skill installation

Direct skill installation is useful for local experimentation:

$skill-installer install https://github.com/dd3ok/cloud-decision-kit/tree/main/plugins/cloud-decision-kit/skills/cloud-decision-kit

For repository-scoped use, copy or symlink the runtime skill to:

<repository>/.agents/skills/cloud-decision-kit/

Usage

Invoke the skill explicitly with $cloud-decision-kit.

Use $cloud-decision-kit to compare the best architecture for this workload.

- Kotlin Spring Boot API
- PostgreSQL and Redis
- Users primarily in Korea; Seoul or Tokyo required
- Initial 30 RPS, peak 500 RPS
- 2 TB monthly internet egress
- PITR required
- One operator

Compare Low, Base, and Peak cost, 12-month TCO, SLA, quotas,
backup, operational burden, lock-in, and the threshold that
would reverse the recommendation.

More example prompts:

Use $cloud-decision-kit to compare managed game backend options
for 2,000 base CCU and 12,000 peak CCU, including session hosting,
matchmaking, relay, voice, warm capacity, and regional coverage.
Use $cloud-decision-kit to compare an AI API architecture with
self-hosted GPU inference. Include utilization sensitivity,
idle GPU cost, scaling limits, data policy, and the break-even point.

Cloud Decision Kit may also activate implicitly when a request clearly asks for cloud provider, architecture, pricing, TCO, or policy comparison.

What a result includes

A Standard or Deep comparison normally contains:

  • workload assumptions and hard gates;
  • eliminated candidates and reasons;
  • one primary recommendation and one runner-up;
  • Low, Base, and Peak monthly cost;
  • 12-month TCO and excluded or unknown charges;
  • source-linked material facts with checked dates;
  • operations, reliability, policy, and lock-in risks;
  • confidence levels and unverified items;
  • sensitivity analysis and reversal thresholds;
  • measurements to collect before committing.

Bundled tools

The runtime skill includes standard-library-only Python tools:

cd plugins/cloud-decision-kit/skills/cloud-decision-kit

# Find relevant official pricing and policy sources.
python scripts/find_sources.py `
  --providers cloudflare `
  --services workers `
  --format markdown

# Validate the source registry.
python scripts/validate_sources.py --json

# Inspect the deterministic cost-model input format.
python scripts/cost_model.py --example

find_sources.py returns discovery URLs; it does not treat the registry as cached truth. cost_model.py calculates supplied inputs and never fetches prices.

Evidence policy

Cloud Decision Kit follows a few strict rules:

  • never present an undated price as current;
  • use official provider sources for material facts;
  • label region, currency, plan, and commitment assumptions;
  • separate temporary credits and free tiers from steady-state cost;
  • include egress, backups, observability, support, and idle capacity when material;
  • mark unverified claims and reduce confidence accordingly;
  • prefer ranges over false precision;
  • do not treat provider calculators or marketing summaries as contractual quotes.

Repository structure

.
├── .agents/plugins/marketplace.json
├── .github/workflows/validate.yml
├── plugins/cloud-decision-kit/
│   ├── .codex-plugin/plugin.json
│   ├── LICENSE
│   └── skills/cloud-decision-kit/
│       ├── SKILL.md
│       ├── agents/openai.yaml
│       ├── references/
│       └── scripts/
├── scripts/
│   ├── package_release.py
│   └── validate_repo.py
├── tests/
├── README.md
└── README.ko.md

The runtime skill is the source of truth:

plugins/cloud-decision-kit/skills/cloud-decision-kit/

User documentation and development notes stay at the repository root so the installed skill remains concise.

Development

Requirements:

  • Python 3.11 or newer;
  • no third-party Python runtime packages.

Run the required checks:

python scripts/validate_repo.py
python -m unittest discover -s tests -v

Useful smoke tests:

cd plugins/cloud-decision-kit/skills/cloud-decision-kit
python scripts/validate_sources.py --json
python scripts/find_sources.py --tags game,realtime --format markdown
python scripts/cost_model.py --example

See CONTRIBUTING.md for contribution guidelines.

Release

  1. Update the plugin version and changelog.
  2. Run validation and tests.
  3. Build deterministic release archives.
  4. Commit, tag, and publish the archives as GitHub Release assets.
python scripts/package_release.py

Generated archives are written to dist/ and intentionally ignored by Git. Source files remain the repository source of truth.

Scope and safety

Cloud Decision Kit supports architecture selection and pre-commitment cost analysis. It does not:

  • deploy infrastructure;
  • generate or apply IaC;
  • access cloud accounts or billing data;
  • investigate post-billing anomalies;
  • replace legal, compliance, tax, or contractual review;
  • provide a binding provider quote.

See PRIVACY.md and SECURITY.md for details.

License

MIT License. See LICENSE.

Releases

Packages

Contributors

Languages