Skip to content

fix(opencode): align Sol defaults and reasoning policy (0.5.4) - #129

Merged
xnoto merged 4 commits into
mainfrom
fix/opencode-sol-model-policy
Oct 13, 2026
Merged

xnoto merged 4 commits into
mainfrom
fix/opencode-sol-model-policy

Conversation

@xnoto

@xnoto xnoto commented Oct 13, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Release opencode-server 0.5.4 for the owner-approved model-policy update:

  • Set the global default to openai/gpt-6.1-sol; keep default_agent: default.
  • Upgrade the six existing GPT-6 Sol agents to GPT-6.1 Sol.
  • Configure medium for all ten primary preferences and terra/cloud/DevOps; xhigh for adversarial/security review; high for recruiter; low for Luna; medium for QA.
  • Correct the cloud reviewer's stale Terra sentence and document session-selection precedence and behavioral validation.

OpenCode image/appVersion 2.0.22, API selector 2, 25-agent inventory, non-OpenAI workers, permissions, MCPs, provider/Secret wiring, and opt-in pilot are unchanged.

Type of change

  • Bug fix
  • Feature / enhancement
  • Documentation
  • Infrastructure (OpenTofu root or module)
  • GitOps desired state (packaged Helm configuration)
  • Container image
  • CI / reusable workflow (existing Makefile/test assertions; workflows unchanged)
  • Refactor / cleanup
  • Breaking change

Validation

  • Pull-request detection, test and packaging checks passed in run 38444720840; downstream updater correctly skipped for PR. CI is the validation environment; no local checks claimed.
  • Generated or centrally distributed files remain generated by their owning automation: Helm ConfigMap/mount/archive parity is tested; no generated copies hand-edited.

Authored checks enforce the independent exact eighteen-agent OpenAI model/variant matrix in source and rendered ConfigMap, reject duplicate/missing/wrong selection headers, assert the global default and retained provider wiring, and preserve the fixed historical baseline, exact approved transforms, primary policy tails, pilot isolation, source/render/archive parity, and package version.

Independent static reviews: adversarial ADVANCE, infra-security ADVANCE, QA COVERED, DevOps ADVANCE for the bounded Makefile and unchanged delivery integration, release READY. No Critical/High finding remains. Release review's numeric-version prose was unreliable and is explicitly superseded: the primary verified canonical Chart.yaml 0.5.4, appVersion 2.0.22, Makefile archive opencode-server-0.5.4.tgz, and the consumer assertion directly.

Catalog supports the selected models/variants; this is not account entitlement, effective request effort, inference, tool-call success, quality, billing, or benchmark evidence. Fresh-session web/TUI/API and child-agent behavior remains separately verified after an approved rollout. Primary preferences do not replace persisted/client-selected session models; the root has no effort override.

Impact and rollout

Producer: makeitworkcloud/charts/opencode-server; consumers: packaged OpenCode primary/subagent preferences and makeitworkcloud/kustomize-cluster/workloads/apps/opencode-app.yaml.

  1. Source/tests/docs authored on this branch; PR hygiene, strict Helm render/lint, tests, detection and packaging run automatically. PR does not publish.
  2. Owner explicitly approved merging when ready, after reviews and green required CI. A confirmed merge triggers existing automatic immutable GHCR chart publication.
  3. Existing publisher automatically proposes the pin-only automation/opencode-server-0.5.4 consumer PR with check-gated auto-merge; workflow unchanged.
  4. Known separate consumer gate: at cluster SHA 5d51403d9ec697f0ea020007ad5872cd222fbb95, .github/workflows/test.yml:916 asserts (0.5.3, 2). The generated 0.5.4 pin PR cannot pass until the consumer coordinates the expected chart version to 0.5.4, retaining API selector 2. This PR does not edit cluster desired state or that assertion and does not bypass checks. Publication may complete while selection stays blocked.
  5. GitOps selection, root/child revision submission and reconciliation, pod replacement/health, and functional verification are distinct later stages; none is claimed here. Configuration is copied at startup and rolls via checksum, not hot reload. No manual sync or live mutation is performed.

Rollback is a separately authorized GitOps chart-pin revision; no image downgrade, credential change or database operation is part of this change.

Safety and secrets

  • Changed content contains no plaintext secrets, decrypted SOPS values, state files, kubeconfigs, tokens, or newly disclosed private endpoints.
  • No local OpenTofu init/plan/apply/destroy/import/state operations were run or claimed — plans come from pull-request checks.
  • Effects and rollback boundaries are described above; no irreversible migration is introduced.

AI-assisted change authored by the makeitwork primary, with independent reviewer subagents. The initial implementation worker made no commits; its write authority was withdrawn before the replacement branch was authored.

@github-actions

Copy link
Copy Markdown

Chart CI passed

Repository hygiene and Helm validation passed. View the workflow run.

@xnoto
xnoto merged commit 9d590fb into main Oct 13, 2026
4 checks passed
@xnoto
xnoto deleted the fix/opencode-sol-model-policy branch October 13, 2026 18:14
xnoto added a commit to makeitworkcloud/kustomize-cluster that referenced this pull request Oct 13, 2026
## Summary

Select published opencode-server chart **0.5.4** from [charts PR
#129](makeitworkcloud/charts#129), and
coordinate the existing exact chart/API assertion to **(0.5.4, 2)**.
Exporter API selector remains **2**. No test relaxation, workflow
behavior, image, Secret, RBAC, network or pilot changes.

## Validation

- Producer publication and updater passed [run
38445170845](https://github.com/makeitworkcloud/charts/actions/runs/38445170845).
- Initial consumer test failed its stale chart/API assertion; this
follow-up corrects only the expected chart version.
- Independent adversarial and infra-security reviews ADVANCE; DevOps
ADVANCE explicitly bounded to edited assertion and supplied unchanged
delivery contract, not overall workflow approval. Primary inspected full
workflow and exact two-file diff. No remaining Critical/High finding in
scoped change.
- Required PR test and logging checks passed in [run
38446019338](https://github.com/makeitworkcloud/kustomize-cluster/actions/runs/38446019338)
at `07cfd08613e64ce2505c3afa96e6d7913d970908`. No local checks claimed.

## Impact and rollout

Owner authorized proceeding with consumer coordination. Existing
generated PR auto-merge is enabled and check-gated. A merge runs main
CI; successful main automatically requests root sync at the tested SHA,
then gitops-workloads updates the opencode child. Root/child revision
reconciliation, resource/pod health, delivered configuration and runtime
behavior remain separately verified; no manual sync or live patch is
performed. Recreate rollout may briefly interrupt the single-replica
server. Image/appVersion 2.0.22 unchanged. Stored/client-selected
sessions retain model precedence; new model entitlement, effective
effort, inference/tool-call success and comparative quality/cost are not
established by static CI.

Rollback requires a reviewed GitOps pin and matching exact assertion to
a published prior version; no retagging, image downgrade or database
operation.

## Safety

No plaintext secrets, decrypted values, credentials, state or
kubeconfigs retrieved or changed. AI-assisted scoped follow-up;
generated chart pin remains publisher-owned.

---------

Co-authored-by: makeitworkbot[bot] <321979582+makeitworkbot[bot]@users.noreply.github.com>
Co-authored-by: Steven Welch <steven@makeitwork.cloud>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant