{
  "surface": "web",
  "viewportCss": {
    "width": 1280,
    "height": 800
  },
  "theme": "light",
  "provider": {
    "source": "ACP Registry",
    "agentId": "mistral-vibe",
    "version": "3.1.1",
    "distribution": "binary",
    "environment": {
      "MISTRAL_API_KEY": "dummy-key-for-model-metadata-proof",
      "VIBE_MODELS": "[{\"name\":\"mistral-vibe-cli-latest\",\"provider\":\"mistral\",\"alias\":\"mistral-medium-3.5\",\"display_name\":\"Mistral Medium 3.5\",\"thinking\":\"high\",\"thinking_levels\":[\"off\",\"high\"]},{\"name\":\"glm-5.3-flash\",\"provider\":\"mistral\",\"alias\":\"glm-5.3-flash\",\"display_name\":\"GLM 5.3 Flash\",\"thinking\":\"high\",\"thinking_levels\":[\"low\",\"high\",\"max\"]}]"
    }
  },
  "setup": [
    "Start each commit in its own worktree with T3CODE_DEV_SEED=0 and fresh state.",
    "Add Mistral Vibe from Settings > Providers > Add provider > Show more > ACP Registry.",
    "Continue to sign-in, skip sign-in, and set the two provider environment variables above.",
    "Refresh provider status. Open the fresh default thread in the synthetic server project. Hide thread details to keep the environment label out of captures.",
    "Open model picker, choose Mistral Vibe, then Mistral Medium 3.5. Its Thinking menu has Off/High."
  ],
  "interaction": [
    "Start preview recording with Medium selected and picker open.",
    "Escape; type Say hi into the composer; submit.",
    "Wait for the invalid-API-key provider error.",
    "Open picker; select GLM 5.3 Flash; leave picker open and capture its options before a GLM session.",
    "Escape; type Say hi again; submit.",
    "Wait for the rejected-credential provider error.",
    "Open picker; head additionally captures learned GLM Low/High/Max.",
    "Select Mistral Medium 3.5; leave picker open and capture its Thinking levels.",
    "Stop recording."
  ],
  "execution": "These actions were issued sequentially through T3 preview tools on each commit. This manifest records the driver inputs; it is not a scenario test.",
  "observations": {
    "base": "Unlearned GLM Off/High; Medium after GLM Low/High/Max.",
    "head": "Unlearned GLM no Thinking choices; learned GLM Low/High/Max; Medium after GLM Off/High."
  },
  "limitations": [
    "Dummy credentials cannot observe an effective reasoning level during successful generation.",
    "Native VIBE_MODELS configuration supplies both model definitions because the dummy-key catalog exposes only Medium by default. No T3 provider snapshot or ACP messages were mocked."
  ]
}
