Source register
What this guide is built on.
Checked on September 13, 2026. Product behavior, model names, plugin availability, and skill locations can change; use these links as the refresh points.
Codex and repository guidance
Custom instructions with AGENTS.md
Official Codex guidance for discovery, precedence, global/project layering, nested overrides, code review rules, fallback names, and verification of loaded instructions.
Harness engineering: leveraging Codex
OpenAI’s 2026 engineering report on repository knowledge as a system of record, progressive disclosure, agent legibility, mechanical invariants, and feedback loops.
Codex use cases
Current examples of coding, research, testing, security, deployment, documentation, data, and long-running workflows.
AGENTS.md
The open, cross-tool convention for repository context, commands, testing, style, security, and nested project guidance.
Skills and reusable workflows
Build skills
Official guidance on skill structure, progressive disclosure, invocation, local discovery locations, plugins, optional metadata, installation, and skill best practices.
Agent Skills open standard
The portable standard that OpenAI’s skill guidance builds on.
OpenAI Plugins
Current official Codex plugin examples with skill-backed workflows. The older openai/skills repository identifies itself as deprecated; use it only as historical context.
OpenAI skills catalog
Historical Codex skill catalog retained for context; follow its notice toward current plugin examples and check each license.
Skills API reference
API-level information about listing, creating, versioning, updating, deleting, and downloading skills.
Prompting, testing, and agent design
OpenAI model guidance and prompting
Current guidance on outcome-first prompts, success criteria, stopping conditions, reasoning effort, verbosity, tool descriptions, structured outputs, compaction, and representative evaluation.
Building effective agents
Anthropic’s practical distinction between workflows and agents, and its advice to use the simplest effective architecture.
Trustworthy agents in practice
A perspective on human control, alignment with expectations, security, transparency, and privacy.
NIST AI Risk Management Framework
A voluntary framework for trustworthy AI risk management, including govern, map, measure, and manage across the AI lifecycle. NIST notes that the framework is being revised, so check the source for current status.
Media generation and model paths
Fooocus
Official repository for the offline, open-source, SDXL-based image UI. Its project status says long-term support is limited to bug fixes, so check the repository before starting a new dependency.
ComfyUI
Official modular node and graph interface for local and cloud workflows across image and other media ecosystems.
Stable Diffusion XL 1.0 model card
Primary model card describing SDXL base and refiner use, hardware notes, limitations, and the model license.
Black Forest Labs FLUX.2 overview
Official overview for current FLUX.2 variants. Verify the exact checkpoint and license rather than treating “open weights” as one universal permission.
BFL self-hosted commercial terms
Primary legal terms for commercial self-hosting decisions, including licensing and operational requirements.
Hosting and rented capacity
Hetzner Cloud server overview
Official documentation for Cloud servers, firewalls, volumes, locations, and the responsibilities that remain with the operator.
Hetzner GPU servers
Current GPU server documentation. Availability and hardware change, so use the live page when comparing a purchase or rental.
Hetzner price adjustments
Pricing is volatile. Treat current provider pages and the checkout view as authoritative instead of copying a stale number into this handbook.
RunPod Pod pricing
Official pricing documentation for per-second Pod billing, storage costs, savings options, and the warning that Pods are not long-term storage.
Vast.ai
Official marketplace entry point for comparing and provisioning community-provided GPU capacity. Inspect the specific host, reliability, storage, and security conditions.
Multi-agent orchestration
OpenAI Agents SDK: multi-agent
Official patterns for manager agents-as-tools, handoffs, code orchestration, parallel work, specialization, and evaluation.
OpenAI Agents SDK: handoffs
Official guidance on transferring ownership, routing context, input schemas, filtering, and keeping authorization before side effects.
OpenAI Agents SDK: agents
Reference for agents, tools, handoffs, managers, lifecycle events, and the basic building blocks of an agent environment.
Harness engineering
OpenAI’s engineering account of making repositories legible, mechanically checkable, and suitable for reliable agent work.
How to use this page
Follow the local guide for practical habits. Follow the source links for exact product behavior, version-sensitive details, legal or organizational decisions, and anything consequential. If a source disagrees with this guide, the source wins and the guide should be updated.