v1.158.0 cannot parse config files it writes — "yaml: line 2228: did not find expected key"

[dfinley]

Summary

doctl v1.158.0 (released 2026-05-07 16:21 UTC) cannot read its own config files. Every doctl invocation fails at startup with:

doctl: 2026/05/07 21:08:20 Config initialization failed: While parsing config: yaml: line 2228: did not find expected key

v1.157.0 reads the same file successfully — verified.

Reproduction

1. Install doctl v1.158.0
2. Run any doctl command (auth init, auth list, registry login, etc.)
3. The first invocation may write a default config (37,883 bytes / 2,227 lines on Linux x86_64); subsequent invocations fail to read it.

The error appears at line 2228 of a 2227-line file — i.e., doctl's parser expects more YAML content than the writer produced. The file ends mid-section with vpcs.update.name: "".

Evidence the file itself is valid YAML

yaml.safe_load (PyYAML) parses the file successfully:

• 91 top-level keys
• Well-formed structure (1-click, access-token, account, action, activations, agent, alert, api-url, apikeys, apps, ...)

So the failure is in doctl's parser, not the file.

Evidence v1.157.0 reads the same file

$ wget https://github.com/digitalocean/doctl/releases/download/v1.157.0/doctl-1.157.0-linux-amd64.tar.gz
$ tar xf doctl-1.157.0-linux-amd64.tar.gz
$ ./doctl --config=/home/USER/.config/doctl/config.yaml auth list
default (current)        # <-- success on the same file v1.158.0 fails on

Suspected commit

The v1.157.0 → v1.158.0 changelog has only 2 commits:

• #1835 DOKS SSO: per-cluster config, kubeconfig type — explicitly modifies config schema structure
• #1839 Bump godo to v1.189.0 — internal HTTP client only, unlikely cause

#1835 is the prime suspect — config schema additions can produce serialization output that the parser doesn't expect.

Impact

• All CI runs using digitalocean/action-doctl@v2 (which defaults to version: 'latest') broke today within hours of the v1.158.0 release.
• Workaround: pin to version: '1.157.0' in workflow input.
• Long-term users with shared doctl config homes (e.g., self-hosted runners) are affected even after pinning, since the broken file persists until manually cleared. v1.157.0 reads the v1.158.0-written file fine, so the workaround does not require config cleanup.

Environment

• doctl: v1.158.0
• OS: Ubuntu (self-hosted GitHub Actions runner, Linux x86_64)
• Action: digitalocean/action-doctl@v2 with version: 'latest' (default)

Workaround applied on our side

Pinning version: '1.157.0' in all digitalocean/action-doctl@v2 callsites until v1.159.0 (or v1.158.x) ships a fix.

Happy to provide the corrupted config file (with secrets redacted) on request.