workflow-migrate atlas driver: panic on apply (Executor.Execute, index out of range [0] with length 0)

[intel352]

Severity: HIGH — blocks atlas-driver upgrade flows

Source

Surfaced during core-dump P2 PR #149 round 2 CI run. Attempted `workflow-migrate up --driver atlas` against an ephemeral postgres-with-AGE container with the merged 232-file migrations corpus.

Symptom

```
panic: runtime error: index out of range [0] with length 0

goroutine 1 [running]:
ariga.io/atlas/sql/migrate.(*Executor).Execute(...)
```

Scope

• Reproducible: yes (CI run https://github.com/GoCodeAlone/core-dump/actions/runs/25214000275)
• Image: `ghcr.io/gocodealone/workflow-migrate:v0.3.6@sha256:b10ba85a...`
• Driver: `atlas`
• Same image + same driver, `down` subcommand WORKS (Gate 5 of migrations-ci.yml succeeds). So the bug is scoped to the apply (`up`) code path of the atlas executor.

Workaround

Gates 3 + 4 in core-dump's migrations-ci.yml are demoted to `continue-on-error: true` (advisory) until this is fixed. Once upstream lands a fix, the gates can be promoted to blocking via a small workflow tweak.

Suggested investigation

• The panic location is `ariga.io/atlas/sql/migrate.(*Executor).Execute`. Look at what slice/array that function indexes at offset [0].
• Suspect: revisions-table inspection or migration-set-by-version lookup that returns an empty slice when the table is freshly created.
• Test: run `up` against a brand-new empty PG18 (no atlas_schema_revisions) — does the panic reproduce on the FIRST migration apply?

Filed

2026-05-01 from core-dump P2 implementer's round-2 investigation.