[codex] run brain deployments embedded

[Che-Zhu]

Summary

• Refactor Brain deployments to run as local embedded Codex tasks instead of creating Devbox or connecting to an inner gateway.
• Make Brain deployment a polling-only API that installs the deployment skill, builds/pushes to GHCR, and returns the final machine-readable result.
• Include full Sealos template YAML content in successful deployment status responses, and harden result parsing against stale assistant messages and template-path false positives.

Behavior

• POST /api/brain/deployments starts a deployment task in the current gateway runtime.
• GET /api/brain/deployments/:threadId returns running, succeeded, or failed.
• Success requires a valid ghcr.io/... image and valid Sealos template content in template; running/failed responses return image: null and template: null.
• The gateway no longer owns Devbox lifecycle for Brain deployment requests.

Validation

• cargo fmt --check
• cargo test
• Two-pass subagent review; initial blockers fixed and second-pass review found no critical or important issues.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 881c0a2c3b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Keep deployment runnable when skill is already preinstalled

The new prompt unconditionally requires npx --yes skills add ... as the first step and explicitly tells the agent to fail the deployment if that install command fails. Before this change, Devbox-backed sessions could skip installation when the skill was prebootstrapped. In environments where outbound npm/GitHub access is restricted but the skill is already present (a common hardened Devbox setup), deployments that previously succeeded will now deterministically fail at startup.

Useful? React with 👍 / 👎.