P1: Downstream canary CI: build stella and the oxagen copy against HEAD to catch breaking changes before the freeze

[macanderson]

Context

Two known downstreams consume this code today: stella (git-pinned ocp-types/ocp-host — see stella#277 for their repin) and oxagen's vendored copy (oxagen-platform#1082). The spec is pre-freeze and changing fast (#4–#13); nothing tells us when a change breaks a real consumer until they bump their pin weeks later.

Task

A scheduled/on-PR CI job that checks out stella and builds its contextgraph-consuming crates against this repo's HEAD (cargo patch/override), reporting breakage as a signal — advisory, not blocking, so the spec can still move. Add oxagen once #1082 lands them on the canonical crates.

Acceptance

• A breaking API change here produces a visible canary failure naming the downstream, before any downstream files a bug.

[macanderson]

This canary is the code-side half of the #27 boundary enforcement

The #27 reconciliation + ADR 0007 make CGP the single normative home for frame/wire semantics; downstream docs (stella, oxagen) now hold only a pinned pointer. Enforcement is two halves:

• Docs side (shipped): each downstream restatement doc carries a NORMATIVE-HOME: macanderson/context-graph-protocol @ <sha> header. stella wires a drift guard (scripts/check-normative-home.sh + a normative-home workflow, stella#500) that fails if the doc pin diverges from the contextgraph-* rev stella builds against. So the docs can't silently re-drift from the code.
• Code side (this issue): the canary should build stella and the oxagen copy against CGP HEAD, so a breaking change to the atomic ContextFrame / wire types is caught before the freeze. That's the guard the pointer-only docs rely on — with semantics living in one place, the only remaining drift risk is a code/type break, which is exactly what this canary catches.

Concretely, the canary wants: (1) cargo build/cargo test stella-graph + stella-context against CGP HEAD (they consume contextgraph-types::ContextFrame directly); (2) validate oxagen's pinned CGP conformance fixtures (@oxagen/run-evidence, ADR-035) against HEAD. Sequenced after the pre-freeze tag (#30) so there's a stable base to diff from.

[macanderson]

Re-verified 2026-07-26 against origin/main @ 1ee9b1c.

One workflow file exists, and there are zero hits for stella, oxagen, schedule: or cron anywhere under .github/ — the canary does not exist in any form. Meanwhile eight wire-affecting PRs have landed since this was filed, which is exactly the churn the canary was meant to observe, and it observed none of it.

Recommended scope: keep as-is; nothing in the body has gone stale. Optionally fold the comment thread's concrete target list into the body so the issue is executable without reading the thread first.