Nothing consumes registry.westgate.pw/deployment/crucible: prod.yml no longer reserves a slot for it and no runtime or recovery path pulls it. Binaries, wrappers, and the Nix input are the supported ways to run Crucible. Delete the image workflows (build/publish/test), the Dockerfile, and the make image target; update docs to match. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2.8 KiB
Crucible consumer contract
How the artifact repos (sow-module, sow-topdata, sow-assets-manifest)
locate and invoke a Crucible builder. The goal: no local-machine assumptions
(Phase 5 exit criterion). A consumer either finds a Crucible binary or fails
closed — it never fakes an artifact.
Resolution order (single-token command)
Each consumer wrapper resolves its builder to one executable token, in order:
- Explicit override —
$SOW_MODULE_BUILD/$SOW_TOPDATA_BUILD/$CRUCIBLE(a path or name), if set and executable. Back-compat with the pre-Crucible skeletons. - Standalone Crucible binary —
crucible-module/crucible-topdata/crucible-hak/crucible-depotonPATH. - Legacy name —
sow-module-build/sow-topdata-buildonPATH(pre-D11; kept so nothing breaks during cutover).
If none resolve, the wrapper exits non-zero with a Phase 5 message. The
crucible dispatcher (crucible module …) is for humans and CI introspection;
wrappers prefer the single-token crucible-<name> shim so "$builder" args
quoting stays correct.
Parity contract — wrappers are not authoritative for build inputs
A consumer build wrapper may only:
- validate — a pre-build gate that does not change output;
- publish — a post-build step on a separate artifact;
- resolve the crucible binary — the bootstrap above.
A wrapper MUST NOT fetch, resolve, generate, or stage any input Crucible reads to
produce the artifact. If Crucible needs an input, Crucible acquires it. This makes
the "no local-machine assumptions" + "same inputs → identical output" rules
explicit: local crucible <build> and CI crucible <build> against the same
committed config must produce identical bytes, with no CI-only pre-steps.
Builders read every input from the project tree. No NWN install is required. A builder that one day needs NWN game data auto-detects the install (the NWN home and Steam path are reliably discoverable), with an optional explicit override for non-standard layouts — never a required hand-set env.
In CI
Builds run the Crucible binary — from the crucible.sh / crucible.ps1
bootstrap (anonymous download) or, for Nix users, from the flake devshell pinned
by flake.lock. CI runs the same crucible <build> as local dev, inside the
nix devshell (binary-cache fast). No build container, no token, no CI-only
pre-steps.
The registry.westgate.pw/deployment/crucible:<sha> image is retired
(2026-07-17): nothing consumed it, prod.yml no longer reserves a slot for
it, and CI no longer builds or publishes it. Binaries, wrappers, and the Nix
input are the only supported ways to run Crucible.
Determinism
Same inputs → identical output bytes. Consumers may checksum artifacts across runs; non-determinism is a builder bug.