project plan, values, milestone 1 spec, zig 0.16 api research

This commit is contained in:
2026-07-31 23:16:57 +02:00
commit b609176a0b
4 changed files with 1148 additions and 0 deletions
+37
View File
@@ -0,0 +1,37 @@
# AGENTS.md
## Aim
nxdns: a self-hosted DNS sinkhole for a household LAN, written in Zig 0.16.0.
Portfolio-grade public repo. PLAN.md is the source of truth for scope and design;
specs/ holds per-milestone contracts; specs/research/ holds verified stdlib facts.
## Values
We intentionally architect this code to be robust, maintainable, pragmatic —
good craftsmanship and good engineering. We explicitly avoid tech debt, code
smells, bad architecture decisions, and brittle implementations.
What that means in practice:
- This is a greenfield project. Breaking changes are allowed. Never keep a bad
interface for compatibility; fix it at the root.
- No versioning of scope. A feature is in scope (build it completely) or out of
scope (do not build it). No "v2 later", no stubs left behind.
- Fix root causes, not symptoms. Do not iterate on workarounds.
- Scope is small on purpose: household scale, two targets, few dependencies.
Do not add generality nobody asked for.
- Dependencies are liabilities: stdlib first; vendored + pinned C deps
(sqlite3, mbedTLS) only where the stdlib has nothing.
- Verify stdlib claims against ../zig at tag 0.16.0 — pre-0.16 knowledge is
stale (std.Io migration). See specs/research/zig-0.16-api-notes.md.
- Pure core: dns/, filter/, local/, cache/ take bytes and return bytes — no Io,
no sockets, no clocks hidden inside.
- Every failure mode must be visible: no silent drops, no unbounded logs, no
swallowed errors. Counters + health surfaces over log spam.
- Tests are runnable acceptance criteria, not decoration. Required CI stays
deterministic — no network-dependent tests in blocking jobs.
- Comments state constraints the code cannot show. No narration, no
commented-out code.
- Git: GPG-signed commits (`git commit -S`), simple lowercase messages, no
generated-by footers.