Downstream Sync

The mirror of /upstream-sync: walk this fork's kit-generic surfaces, find changes the canonical kit should adopt (bug fixes in scaffold code, hardened checks, new reusable components, workflow improvements), and author contribution kits — one spec file per candidate. The deliverable is a spec, not a patch: a spec the kit's own assistant can execute travels better than a diff that assumes this fork's layout.

Pre-flight

1. Self-detection. pnpm kit:check --why reports canonical → print downstream-sync: canonical kit repo — skipping (N/A). and stop.
2. State. Read kit.json → sync.downstream (lastCheckedDate, proposed[]) and kit.json → paths.appSpecific (this fork's product surfaces — never candidates). paths.personalized entries are identity, not contributions.

Known Untested Paths

Same caveat as upstream-sync: structurally verified, never run against a live fork. First fork owner to run this is the first real test.

Step 1 — Enumerate kit-generic surfaces

Walk, in this order (skip anything under paths.appSpecific or paths.personalized):

• scripts/ (tripwires, kit tooling) and .github/workflows/
• packages/{auth,db,permissions,ui,tokens,brand,config}/ — generic infrastructure only
• apps/*/src/proxy.ts, shared components under apps/*/src/components/shared/
• .claude/skills/ and .claude/agents/
• AGENTS.md, UI-STANDARDS.md, UX-PATTERNS.md, BRANDING.md — generic sections

For each, diff against the baseline (git log <sync.upstream.baselineSha>.. -- <path> when history is shared; otherwise reason from the fork's own commit history) and collect changes that are generic — they would help every fork, not just this product.

Step 2 — Classify candidates

Step 3 — Author contribution kits

For each non-skip candidate, write docs/starter-contributions/NN-<slug>.md:

# NN — <title>
Origin: <fork commit(s) / work-log>
What & why: <the problem and the fix, 3–8 sentences>
Applies to the kit as: <where it lands in the kit's layout>
Implementation steps: <numbered, executable by the kit's assistant>
Verification: <how the kit proves it works>
Classification: <class> · Risk: <low/medium/high + why>

Number sequentially from the highest existing NN. Submit by PR into the canonical repo's docs/starter-contributions/incoming/ — this skill never opens the PR; it tells the user the exact gh pr create invocation to run if they want it.

Step 4 — Log + state

• Append YYYY-MM-DD | downstream-sync | N candidates (A ready, B needs-gen, C structural) (or nothing to contribute) to docs/reviews/log.md.
• Update kit.json.sync.downstream: lastCheckedDate = today; add each authored candidate to proposed[] as {slug, date, class} so the next run skips it unless the underlying code changed again.
• Failures before Step 1 completes: write nothing.