Run SDK evolution update across vendor packages

[ebarti]

SDK Evolution Report

SDK Evolution Agent Report

Run

• Runtime: antigravity-agent-sdk
• Implementation enabled: True
• Draft PR enabled: True

Upstream Evidence

• claude-agent-sdk: locked=0.2.96 installed=0.2.106 latest=0.2.106
• openai-codex: locked=0.1.0b3 installed=0.1.0b3 latest=0.1.0b3
• openai-codex-cli-bin: locked=0.137.0a4 installed=0.137.0a4 latest=0.136.0
• google-antigravity: locked=0.1.2 installed=0.1.4 latest=0.1.4

API Diffs

• Diff count: 2

Release Notes

• claude-agent-sdk: found (0.2.96 -> 0.2.106)
• google-antigravity: found (0.1.2 -> 0.1.4)

Behavior Probes

• Status: pass
• Changed contracts: 0
• Breaking contracts: 0
• Diff count: 2

Direction Of Travel

{
  "packages": [
    {
      "direction": "Upgrade to version 0.2.106, adding task status tracking symbols ('TERMINAL_TASK_STATUSES', 'TaskUpdatedMessage', 'TaskUpdatedStatus').",
      "evidence": [
        "Added TERMINAL_TASK_STATUSES, TaskUpdatedMessage, and TaskUpdatedStatus to API.",
        "adapter-contract probe passes successfully for candidate version 0.2.106."
      ],
      "name": "claude-agent-sdk"
    },
    {
      "direction": "Refactor model configuration and capabilities to support multimodal structures (Audio, Video, Image, Document) and unified model endpoints (Gemini, Vertex) while removing obsolete configuration classes (GeminiConfig, ModelConfig, ModelEntry, mcp).",
      "evidence": [
        "Added Audio, Video, Image, Document, ModelTarget, ModelType, VertexEndpoint, GeminiAPIEndpoint to API.",
        "Removed GeminiConfig, GenerationConfig, ModelConfig, ModelEntry, and mcp from API.",
        "Changed CapabilitiesConfig, LocalAgentConfig, and ToolContext configurations."
      ],
      "name": "google-antigravity"
    }
  ],
  "themes": [
    {
      "name": "Multimodal and Unified APIs in Google Antigravity",
      "summary": "Google Antigravity is moving towards a standardized multimodal representation (Audio, Video, Document, etc.) and centralizing endpoint configurations under distinct Gemini and Vertex definitions, streamlining configuration components."
    },
    {
      "name": "Robust Task Lifecycle Tracking in Claude Agent SDK",
      "summary": "The Claude Agent SDK is introducing explicit state machinery elements like TERMINAL_TASK_STATUSES and TaskUpdatedMessage to improve tracking of background or long-running tasks."
    }
  ],
  "uncertainty": [
    "The potential impact on existing client setups from the complete removal of the 'mcp' configuration in google-antigravity.",
    "The exact schema changes to LocalAgentConfig, CapabilitiesConfig, and ToolContext."
  ]
}

Architecture Decision

• Manual design required: False
• Recursive self-adaptation impact: False
• Safe to implement: True

{
  "findings": [
    {
      "classification": "compatible",
      "evidence": [
        "claude-agent-sdk v0.2.96 -> v0.2.106",
        "Probe adapter-contract has status pass with severity none",
        "Added symbols: TERMINAL_TASK_STATUSES, TaskUpdatedMessage, TaskUpdatedStatus"
      ],
      "summary": "Claude Agent SDK introduces background task monitoring symbols with zero contract disruption."
    },
    {
      "classification": "compatible",
      "evidence": [
        "google-antigravity v0.1.2 -> v0.1.4",
        "Probe adapter-contract has status pass with severity none",
        "Removed optional symbols GeminiConfig, GenerationConfig, ModelConfig, ModelEntry, and mcp",
        "Preserved all required LocalAgentConfig adapter fields"
      ],
      "summary": "Google Antigravity restructures config symbols for multimodal endpoints while maintaining core adapter compatibility."
    }
  ],
  "manual_design_required": false,
  "recursive_self_adaptation_impact": false,
  "safe_to_implement": true,
  "self_adaptation_plan": [
    "Upgrade claude-agent-sdk to v0.2.106 and google-antigravity to v0.1.4 in pyproject.toml and update uv lockfile."
  ],
  "uncertainty": [],
  "verification_commands": [
    "uv lock --dry-run"
  ]
}

Implementation Summary

{
  "allowed": true,
  "applied": true,
  "blocked_reason": "",
  "changes": [
    "Updated uv.lock for resolver-selected SDK packages: claude-agent-sdk, google-antigravity"
  ],
  "planned_changes": [
    "Upgrade claude-agent-sdk to v0.2.106 and google-antigravity to v0.1.4 in pyproject.toml and update uv lockfile."
  ],
  "verification_results": [
    {
      "command": [
        "git",
        "switch",
        "-c",
        "sdk-evolution-autonomous-update-20260622"
      ],
      "removed_env": [],
      "returncode": 0,
      "stderr": "Switched to a new branch 'sdk-evolution-autonomous-update-20260622'\n",
      "stdout": ""
    },
    {
      "command": [
        "uv",
        "lock",
        "-P",
        "claude-agent-sdk",
        "-P",
        "google-antigravity"
      ],
      "removed_env": [],
      "returncode": 0,
      "stderr": "Resolved 69 packages in 102ms\nUpdated claude-agent-sdk v0.2.96 -> v0.2.106\nUpdated google-antigravity v0.1.2 -> v0.1.4\n",
      "stdout": ""
    },
    {
      "command": [
        "uv",
        "run",
        "ruff",
        "check",
        "."
      ],
      "removed_env": [],
      "returncode": 0,
      "stderr": "",
      "stdout": "All checks passed!\n"
    },
    {
      "command": [
        "uv",
        "run",
        "mypy"
      ],
      "removed_env": [],
      "returncode": 0,
      "stderr": "",
      "stdout": "Success: no issues found in 14 source files\n"
    },
    {
      "command": [
        "uv",
        "run",
        "pytest"
      ],
      "removed_env": [],
      "returncode": 0,
      "stderr": "",
      "stdout": "============================= test session starts ==============================\nplatform darwin -- Python 3.10.13, pytest-9.0.3, pluggy-1.6.0\nrootdir: /private/tmp/ark-evolution-all-packages\nconfigfile: pyproject.toml\nplugins: asyncio-1.4.0, anyio-4.13.0\nasyncio: mode=strict, debug=False, asyncio_default_fixture_loop_scope=None, asyncio_default_test_loop_scope=function\ncollected 114 items\n\ntests/test_antigravity_adapter.py ..................                     [ 15%]\ntests/test_claude_adapter.py .................                           [ 30%]\ntests/test_codex_adapter.py .......................                      [ 50%]\ntests/test_core.py ....                                                  [ 54%]\ntests/test_events.py .....                                               [ 58%]\ntests/test_live_smoke.py sss                                             [ 61%]\ntests/test_mestre_compatibility.py ..                                    [ 63%]\ntests/test_optional_dependencies.py .                                    [ 64%]\ntests/test_provider_diagnostics.py .                                     [ 64%]\ntests/test_sdk_contract.py .........                                     [ 72%]\ntests/test_sdk_evolution_agent.py ...............................        [100%]\n\n======================== 111 passed, 3 skipped in 0.99s ========================\n"
    },
    {
      "command": [
        "uv",
        "lock",
        "--check"
      ],
      "removed_env": [],
      "returncode": 0,
      "stderr": "Resolved 69 packages in 3ms\n",
      "stdout": ""
    }
  ]
}

Current State Baseline

• Promotion status: promoted
• Promoted: True

Reviewer Output

{
  "reasons": [
    "Adapter-contract probes pass for both packages with severity none.",
    "Required adapter configuration fields are successfully preserved.",
    "API symbol churn does not break the adapter contract."
  ],
  "required_changes": [],
  "status": "pass"
}

Manual Review Checklist

• Verify source references are enough for every architecture finding.
• Verify vendor-specific behavior has not been flattened.
• Verify recursive self-adaptation impact is handled or explicitly blocked.
• Verify tests, docs, examples, and migration notes match public API changes.
• Confirm no auto-merge or unsupported credential scraping was used.

Safety

• Draft PR only.
• No auto-merge.
• Local credentials are not scraped.

[ebarti]

Superseded by #14 after fixing final report handling in the SDK evolution agent.