CodingScaffold Wiki
CodingScaffold is a local-first onboarding, configuration, and governance scaffold for AI-assisted software development teams. The README is the quick front door; this wiki explains the project in more depth and gives teams a shared rollout playbook.
What CodingScaffold Does
CodingScaffold creates project-local guidance and lightweight configuration for:
- local-first model guidance
- provider and credential discovery
- prompt-based model selection
- OpenCode, Claude Code, Codex, OpenClaude, Hermes, and Pi adapters
- reusable team skills
- agent orchestration profiles
- shared Markdown, Obsidian, or MemPalace-ready knowledge bases
- experienced-team onboarding manifests
- policy packs for provider, sharing, permission, and MCP defaults
- context budgeting and optional compression sidecars
- optional RouteLLM and Open Multi-Agent workflows
It does not collect secrets, does not require one model vendor, and does not force a hosted service into the workflow.
It is not a coding agent, a replacement for existing coding tools, an autonomous development platform, a security boundary, or a universal model router.
First-Start Rule
CodingScaffold itself can start without an LLM. Setup, hardware probe, credential templates,
adapter generation, and tools select-model command are local Python workflows. The first actual model
call happens later, inside the coding tool, when a developer runs an agent command such as
/first-session in OpenCode.
Recommended Reading
Start with the smallest path that matches your job today:
Then use the reference pages when the need appears:
- Core Concepts: the vocabulary behind local-first scaffolding.
- Model Selection and Providers: routine vs heavy-lift guidance.
- Skills and Agents: reusable playbooks and agent definitions.
- Context Hygiene: context budgets, linting, and compression sidecars.
- Team Onboarding: manifests for teams that already have shared assets.
- Policy Packs: reviewable provider, sharing, permission, and MCP defaults.
Design Posture
CodingScaffold is intentionally boring where that helps:
- Git is the sharing mechanism.
- Markdown is the source of truth.
- Local credentials stay local.
- Generated files are readable and editable by hand.
- Advanced orchestration is optional and comes after the team validates the workflow.