feat: upstream_fix marker convention, --audit-fixes command, and CI Bun crash resilience (#555)

* feat: add `upstream_fix:` marker convention with `--audit-fixes` command

Introduces a strategy for handling bug fixes to upstream code:

**`upstream_fix:` convention:**
- Tag in marker description distinguishes temporary bug fixes from
  permanent feature additions
- Before each upstream merge, run `--audit-fixes` to review which
  fixes we're carrying and whether upstream has shipped their own

**`--audit-fixes` flag on analyze.ts:**
- Lists all `upstream_fix:` markers with file locations and descriptions
- Prints merge review checklist

**Retagged 3 existing upstream bug fixes:**
- `locale.ts` — days/hours duration swap
- `command/index.ts` — placeholder lexicographic sort
- `home.tsx` — beginner UI race condition

**Code review fixes from #546:**
- Guard `--downstream` without `--model` (returns error instead of
  silently running full project build)
- Remove unused `condition` field from `Suggestion` interface
- Add test for `--downstream` error case

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix: handle Bun segfault during CI test cleanup

Bun 1.3.x has a known crash (segfault/SIGTERM) during process cleanup
after all tests pass successfully. This caused flaky CI failures where
5362 tests pass but the job reports failure due to exit code 143.

The fix captures test output and checks the actual pass/fail summary
instead of relying on Bun's exit code:
- Real test failures (N fail > 0): exit 1
- No test summary at all (Bun crashed before running): exit 1
- All tests pass but Bun crashes during cleanup: emit warning, exit 0

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix: use file redirect instead of tee for Bun crash resilience

The previous approach using `tee` failed because when Bun segfaults,
the pipe breaks and tee doesn't flush output to the file. The grep
then finds no summary and reports "crashed before producing results"
even though all tests passed.

Fix: redirect bun output to file directly (`> file 2>&1 || true`),
then `cat` it for CI log visibility. Use portable `awk` instead of
`grep -oP` for summary extraction.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: anandgupta42

.github/workflows/ci.yml

@@ -84,13 +84,45 @@ jobs:
         run: bun install
 
       - name: Run tests
-        run: bun test --timeout 30000
         working-directory: packages/opencode
         # Cloud E2E tests (Snowflake, BigQuery, Databricks) auto-skip when
         # ALTIMATE_CODE_CONN_* env vars are not set. Docker E2E tests auto-skip
         # when Docker is not available. No exclusion needed — skipIf handles it.
         # --timeout 30000: matches package.json "test" script; prevents 5s default
         # from cutting off tests that run bun install or bootstrap git instances.
+        #
+        # Bun 1.3.x has a known segfault during process cleanup after all tests
+        # pass (exit code 143/SIGTERM or 134/SIGABRT). We capture test output and
+        # check for real failures vs Bun crashes to avoid false CI failures.
+        shell: bash
+        run: |
+          # Redirect bun output to file, then cat it for CI visibility.
+          # This avoids tee/pipe issues where SIGTERM kills tee before flush.
+          bun test --timeout 30000 > /tmp/test-output.txt 2>&1 || true
+          cat /tmp/test-output.txt
+
+          # Extract pass/fail counts from Bun test summary (e.g., " 5362 pass")
+          PASS_COUNT=$(awk '/^ *[0-9]+ pass$/{print $1}' /tmp/test-output.txt || true)
+          FAIL_COUNT=$(awk '/^ *[0-9]+ fail$/{print $1}' /tmp/test-output.txt || true)
+
+          echo ""
+          echo "--- Test Summary ---"
+          echo "pass=${PASS_COUNT:-none} fail=${FAIL_COUNT:-none}"
+
+          # Real test failures — always fail CI
+          if [ -n "$FAIL_COUNT" ] && [ "$FAIL_COUNT" != "0" ]; then
+            echo "::error::$FAIL_COUNT test(s) failed"
+            exit 1
+          fi
+
+          # Tests passed (we have a pass count and zero/no failures)
+          if [ -n "$PASS_COUNT" ] && [ "$PASS_COUNT" -gt 0 ] 2>/dev/null; then
+            exit 0
+          fi
+
+          # No test summary at all — Bun crashed before running tests
+          echo "::error::No test results found in output — Bun may have crashed before running tests"
+          exit 1
 
   # ---------------------------------------------------------------------------
   # Driver E2E tests — only when driver code changes.

packages/dbt-tools/src/commands/build.ts

@@ -2,8 +2,11 @@ import type { DBTProjectIntegrationAdapter, CommandProcessResult } from "@altima
 
 export async function build(adapter: DBTProjectIntegrationAdapter, args: string[]) {
   const model = flag(args, "model")
-  if (!model) return project(adapter)
   const downstream = args.includes("--downstream")
+  if (!model) {
+    if (downstream) return { error: "--downstream requires --model" }
+    return project(adapter)
+  }
   const result = await adapter.unsafeBuildModelImmediately({
     plusOperatorLeft: "",
     modelName: model,

packages/dbt-tools/test/build.test.ts

@@ -45,6 +45,14 @@ describe("build command", () => {
     })
   })
 
+  test("build --downstream without --model returns error", async () => {
+    const adapter = makeAdapter()
+    const result = await build(adapter, ["--downstream"])
+    expect(result).toEqual({ error: "--downstream requires --model" })
+    expect(adapter.unsafeBuildProjectImmediately).not.toHaveBeenCalled()
+    expect(adapter.unsafeBuildModelImmediately).not.toHaveBeenCalled()
+  })
+
   test("build surfaces stderr as error", async () => {
     const adapter = makeAdapter({
       unsafeBuildProjectImmediately: mock(() =>

packages/opencode/src/cli/cmd/tui/routes/home.tsx

@@ -38,7 +38,7 @@ export function Home() {
     return Object.values(sync.data.mcp).filter((x) => x.status === "connected").length
   })
 
-  // altimate_change start — fix race condition: don't show beginner UI until sessions loaded
+  // altimate_change start — upstream_fix: race condition shows beginner UI flash before sessions loaded
   const isFirstTimeUser = createMemo(() => {
     // Don't evaluate until sessions have actually loaded (avoid flash of beginner UI)
     // Return undefined to represent "loading" state

packages/opencode/src/command/index.ts

@@ -54,8 +54,9 @@ export namespace Command {
     const result: string[] = []
     const numbered = template.match(/\$\d+/g)
     if (numbered) {
-      // altimate_change start — fix lexicographic sort of multi-digit placeholders ($10 before $2)
-      for (const match of [...new Set(numbered)].sort((a, b) => parseInt(a.slice(1), 10) - parseInt(b.slice(1), 10))) result.push(match)
+      // altimate_change start — upstream_fix: lexicographic sort of multi-digit placeholders ($10 sorted before $2)
+      for (const match of [...new Set(numbered)].sort((a, b) => parseInt(a.slice(1), 10) - parseInt(b.slice(1), 10)))
+        result.push(match)
       // altimate_change end
     }
     if (template.includes("$ARGUMENTS")) result.push("$ARGUMENTS")

packages/opencode/src/skill/followups.ts

@@ -4,7 +4,6 @@ export namespace SkillFollowups {
     skill: string // skill name to suggest
     label: string // short display label
     description: string // why this is a good next step
-    condition?: string // optional: when this suggestion applies
   }
 
   // Map from skill name to follow-up suggestions
@@ -151,7 +150,8 @@ export namespace SkillFollowups {
   }
 
   // A special warehouse nudge for users who haven't connected yet
-  const WAREHOUSE_NUDGE = "**Tip:** Connect a warehouse to validate against real data. Run `/discover` to auto-detect your connections."
+  const WAREHOUSE_NUDGE =
+    "**Tip:** Connect a warehouse to validate against real data. Run `/discover` to auto-detect your connections."
 
   export function get(skillName: string): readonly Suggestion[] {
     return Object.freeze(FOLLOWUPS[skillName] ?? [])

packages/opencode/src/util/locale.ts

@@ -54,8 +54,10 @@ export namespace Locale {
       const minutes = Math.floor((input % 3600000) / 60000)
       return `${hours}h ${minutes}m`
     }
+    // altimate_change start — upstream_fix: days/hours calculation were swapped (hours used total, not remainder)
     const days = Math.floor(input / 86400000)
     const hours = Math.floor((input % 86400000) / 3600000)
+    // altimate_change end
     return `${days}d ${hours}h`
   }
 

script/upstream/README.md

@@ -222,6 +222,44 @@ When we modify upstream files (not fully custom ones), we wrap our changes with
 
 These help during conflict resolution — you can see exactly what we changed vs upstream code. The `analyze.ts` script audits for unclosed marker blocks.
 
+### Upstream Bug Fixes (`upstream_fix:` tag)
+
+When fixing a **bug in upstream code** (not adding a feature), use the `upstream_fix:` tag in the marker description:
+
+```typescript
+// altimate_change start — upstream_fix: days/hours calculation were swapped
+const days = Math.floor(input / 86400000)
+const hours = Math.floor((input % 86400000) / 3600000)
+// altimate_change end
+```
+
+**Why this matters:** Regular `altimate_change` markers protect features we added — they're permanent. But upstream bug fixes are **temporary**: once upstream ships their own fix, we should drop our marker and accept theirs.
+
+Without the `upstream_fix:` tag:
+- If upstream fixes the same bug, the merge creates a conflict (good — forces review)
+- But the reviewer doesn't know our change was a bug fix vs a feature, so they may keep both
+
+With the `upstream_fix:` tag:
+- Before each merge, run `--audit-fixes` to see all bug fixes we're carrying
+- During conflict resolution, reviewers know to check "did upstream fix this?" and can safely drop our version
+- After merge, any remaining `upstream_fix:` markers represent bugs upstream hasn't fixed yet
+
+**When to use which:**
+
+| Scenario | Marker |
+|----------|--------|
+| New feature/custom code | `// altimate_change start — description` |
+| Fix bug in upstream code | `// altimate_change start — upstream_fix: description` |
+| Branding change | No marker (handled by branding transforms) |
+| Code in `keepOurs` files | No marker needed |
+
+**Audit before merging:**
+
+```bash
+# List all upstream bug fixes we're carrying
+bun run script/upstream/analyze.ts --audit-fixes
+```
+
 ## File Organization
 
 ```

script/upstream/analyze.test.ts

@@ -246,6 +246,18 @@ describe("parseDiffForMarkerWarnings", () => {
     expect(warnings[0].context).toContain("first")
   })
 
+  test("upstream_fix: tagged markers are recognized as valid markers", () => {
+    const diff = makeDiff(
+      `@@ -50,4 +50,6 @@
+ const existing = true
++// altimate_change start — upstream_fix: days/hours were swapped
++const days = Math.floor(input / 86400000)
++// altimate_change end
+ const more = true`,
+    )
+    expect(parseDiffForMarkerWarnings("file.ts", diff)).toEqual([])
+  })
+
   test("real-world scenario: upgrade indicator in footer.tsx", () => {
     // Simulates the exact diff that leaked: UpgradeIndicator added to
     // session footer without markers, adjacent to existing yolo marker block.