# Unified Agent9 Skill Release

<!-- auto-oncall-optimization-record
{
  "schemaVersion": "auto-oncall-optimization-record-v1",
  "optimizationId": "unified-agent9-skill-release-2026-08-21",
  "problems": [
    {
      "problemId": "kubernetes-agent9-skill-source-divergence",
      "summary": "Kubernetes installed Skills from the auto-oncall image while Agent9 Skills were published separately, allowing the two runtimes to use different instructions from the same auto-oncall commit.",
      "status": "optimized_verified",
      "applicability": "Auto-oncall releases whose checked-in Skill packages are consumed by both Kubernetes Codex and the linked Agent9 Agent.",
      "issueKeys": [],
      "sourceRefs": ["manual://auto-oncall-agent9-skill-audit#problem:kubernetes-agent9-skill-source-divergence"],
      "signalTextIncludesAny": [
        "agent9 skills are stale",
        "k8s and agent9 skill sha mismatch",
        "separate skill release lanes"
      ],
      "observedThrough": "2026-08-21",
      "effectiveAt": "2026-08-21T03:38:18Z",
      "verifiedAt": "2026-08-21T03:38:18Z",
      "attemptedAction": "Publish all six checked-in Skill directories to Agent9, advance the linked Agent's exact pins, build one image, and dispatch both staging application bumps from one manual workflow.",
      "evidenceUrl": "https://github.com/tidbcloud/aisupport/actions/runs/32443960915"
    }
  ]
}
-->

- Owner: Auto-oncall staging release workflow
- Source: Kubernetes and Agent9 Skill package comparison
- Implementation status: deployed from `aa103408970e9492873dd53621ab14bb2788ddd3`
- Effective status: `optimized_verified`

## Bounded problem

The same six logical Skills had two delivery lanes. Kubernetes copied the
repository directories into `CODEX_HOME`, while Agent9 retained independently
published immutable versions. A normal source change therefore could update
only one runtime.

## Implemented change

One manual workflow archives the six Skill directories from the exact release
commit, appends or confirms their Agent9 immutable versions, advances the linked
Agent's exact installations, builds one image, and dispatches both staging
application bumps. The server remains the package validator. A failed Agent9
publication stops before Kubernetes deployment dispatch.

Wetech Skill parity is source delivery only. It does not enable Wetech OP; the
support gate and Managed Tool installation remain disabled.

## Verification

The unified release Action completed successfully from exact source commit
`aa103408970e9492873dd53621ab14bb2788ddd3`. It published the six checked-in
Skill packages, advanced the linked Agent9 Agent installations, and dispatched
both staging application bumps. GitOps then converged both active and passive
Kubernetes deployments to that same image SHA with Ready pods and no restart.
The post-deployment Agent9 replay loaded Clinic Skill v4, proving that the
Agent9 and Kubernetes delivery lanes consumed the same release action.