Zig LLM Wiki

A source-driven, LLM-maintained knowledge base for writing robust systems software with Zig 0.16.x.

The wiki is optimized for coding agents first and humans second. It combines current Zig APIs, std.Io semantics, concurrency and cancellation, operating system I/O facilities, and TigerStyle engineering principles. Claims should be easy to locate, hard to misread, and backed by primary sources or executable proofs.

Start here

• Open this directory as an Obsidian vault, then open [[index]].
• Coding agents should read AGENTS.md before consulting or editing the wiki.
• Run zig build verify before accepting content changes.
• Run tools/verify_linux_ssh.sh omarx1 for the complete Linux runtime suite.
• Dispatch Windows runtime verification in GitHub Actions for the exact Windows host/proof suite and retained evidence packet.
• Use python3 tools/wiki.py query ..., lint, plan-only ingest/upgrade, and read-only review as documented in AGENTS.md.
• Use ROADMAP.md as the persistent project plan.
• Use CURATION.md to see which sources are merely discovered, pinned, synthesized, or backed by executable proof.
• Read HANDOFF.md when continuing the project in a fresh agent session; it records exact platform evidence, deliberate decisions, and honest remaining gaps.
• Read the platform testing runbook before changing macOS, Linux, Windows, cancellation, timing, or hosted-verification claims.

The required compiler version is recorded in .zig-version. The initial baseline is Zig 0.16.0, the current stable 0.16.x release as of 2026-09-04.

Repository shape

sources/            pinned source records; evidence is not rewritten
wiki/               maintained synthesis and cross-links
proofs/             executable Zig evidence referenced by wiki pages
benchmarks/          reviewed agent-retrieval regression queries
tools/               query, lint, review, benchmark, and evidence utilities
.github/workflows/   read-only scheduled review packet
.agents/skills/     repo-scoped coding-agent workflow for wiki operations
index.md             content map read before retrieval
log.md               append-only maintenance history
AGENTS.md            schema and maintenance contract
ROADMAP.md           work across content, LLM, and backend lanes
CURATION.md          transparent source-selection and ingestion ledger
HANDOFF.md           durable continuation state and platform evidence matrix
docs/platform-testing.md  macOS/Linux/Windows verification and evidence runbook

Why there is no backend yet

Obsidian already supplies local Markdown browsing, full-text search, backlinks, and an interactive graph. Building a server now would create a second concern before the content model has earned one. The Markdown vault is the product and the source of truth; a future web UI can consume it without changing that.

The decision and the triggers for revisiting it are recorded in ADR 0001.