Ai Configuration
AI Configuration
Claude Code and OpenCode operate at full parity — 5 agents, 5 rules, 7 skills / 9 commands. The configuration ecosystem spans global defaults, project overrides, and extension-triggered rules that auto-load without explicit invocation.
Layer Architecture
Configuration resolves in order: Global → Project → Rules (auto-loaded by extension).
~/.claude/ ├── CLAUDE.md # Behavioral constraints, standards ├── settings.json # Permissions, hooks, env vars ├── rules/ # Auto-loaded by file extension │ ├── asciidoc.md │ ├── bash-shell.md │ ├── d2-diagrams.md │ ├── lua-nvim.md │ └── python.md ├── skills/ # Slash commands (/deploy, /worklog, etc.) ├── agents/ # Specialized AI agents └── hooks/ # ShellCheck, AsciiDoc validation, security
Project-level .claude/ directories override or extend global config.
All global config stowed from dots-quantum via GNU Stow.
Agents (5)
| Agent | Model / Access | Purpose |
|---|---|---|
doc-auditor |
Haiku, read-only |
Audit documentation for standards compliance |
adoc-linter |
Haiku, read-only |
Lint AsciiDoc files for attribute usage, TOC violations, xref correctness |
build-fixer |
Sonnet, full tools |
Diagnose and fix Antora build failures |
project-modularizer |
Sonnet, full tools |
Convert monolithic documents into STD-001 project structure |
reviewer |
Haiku, read-only |
Review content for quality, accuracy, and completeness |
Rules (5)
| Rule File | Scope |
|---|---|
|
Auto-loaded for |
|
Auto-loaded for |
|
Auto-loaded for |
|
Auto-loaded for |
|
Auto-loaded for |
Rules activate automatically when editing matching file types. No explicit invocation required.
Skills (7)
| Skill | Status | Purpose |
|---|---|---|
|
Active |
Deploy spoke repos to Cloudflare Pages via domus-docs |
|
Active |
Create and manage daily worklogs with section partials |
|
Active |
Create collaboration session logs with commits, agents, modes |
|
Active |
Manage projects per STD-001 (create, audit, add-phase, status) |
|
Active |
Scaffold ideas as projects or draft documents |
|
Active |
Save CLI commands to quick-commands codex with context and tags |
|
Active |
Review changed code for reuse, quality, and efficiency |
OpenCode Parity
Nine commands in OpenCode mirror Claude Code skills:
| Command | Purpose | Parity Status |
|---|---|---|
|
Lint AsciiDoc files |
Ported |
|
Run Antora build validation |
Ported |
|
Token usage and cost tracking |
Ported |
|
Spoke deployment |
Ported |
|
Daily worklog management |
Ported |
|
Session log creation |
Ported |
|
Project management |
Ported |
|
Idea scaffolding |
Ported |
|
Command capture |
Ported |
Full parity: 6/6 core skills ported.
Three additional commands (audit-adoc, build-check, cost-report) are OpenCode-native.
Context Library
The context library provides domain-specific knowledge to AI assistants:
-
13 CLAUDE.md files — one per spoke repo, defining component-specific invariants
-
8 context files — domain-specific knowledge (networking, security, infrastructure)
-
3 global baselines — behavioral constraints, CLI coaching, AsciiDoc standards
All context files stored in examples/claude/ for reference and portability.
Governing Standards
| Standard | Scope |
|---|---|
STD-007 |
Delegation Boundary — execution requires explicit grant |
STD-008 |
CLI Quality — verify-before/change/verify-after pattern |
STD-004 |
AsciiDoc — attributes over literals, no inline TOC |
STD-006 |
Secrets — never decrypt, never surface values, document paths only |