From 369f6cb04dcd673293bef33ef737e88c5edb9cdc Mon Sep 17 00:00:00 2001 From: Eloi Date: Sat, 17 Oct 2026 20:34:37 +0200 Subject: [PATCH] Use gpt-6.1-sol at max effort for SDK evolution --- .codex/skills/agent-runtime-kit-upgrade/SKILL.md | 9 +++++++-- docs/sdk-evolution-agent-design.md | 4 ++-- docs/sdk-evolution-agent.md | 8 ++++---- examples/sdk_evolution_agent/stages.py | 4 ++-- tests/test_sdk_evolution_agent.py | 16 +++++++++------- 5 files changed, 24 insertions(+), 17 deletions(-) diff --git a/.codex/skills/agent-runtime-kit-upgrade/SKILL.md b/.codex/skills/agent-runtime-kit-upgrade/SKILL.md index dda102b..af4eeb3 100644 --- a/.codex/skills/agent-runtime-kit-upgrade/SKILL.md +++ b/.codex/skills/agent-runtime-kit-upgrade/SKILL.md @@ -39,7 +39,8 @@ Announce the selected checkout and base. If repository instructions mention a GSD command that is not installed, record the plan with an available planning mechanism and continue the authorized work; do not wait for a missing wrapper. -Use `codex-agent-sdk` unless the user selected another runtime. Match the extra: +Use `codex-agent-sdk` unless the user selected another runtime. Its AI stages +default to `gpt-6.1-sol` with `reasoning_effort=max`. Match the extra: - `claude-agent-sdk` -> `--extra claude` - `codex-agent-sdk` -> `--extra codex` @@ -55,7 +56,7 @@ env -u UV_EXCLUDE_NEWER -u UV_EXCLUDE_NEWER_PACKAGE \ If authentication fails, report the concrete blocker. Do not scrape or reconstruct credentials. Honor an explicitly requested model and reasoning effort rather than silently using the runner's defaults. Pass explicit runner -options such as `--model gpt-6-astra --reasoning-effort xhigh`; `config.json` +options such as `--model gpt-6.1-sol --reasoning-effort max`; `config.json` records them, and each AI stage uses the corresponding first-class task fields. Verify that selection with a small structured runtime call before expensive collection. The installed @@ -74,6 +75,8 @@ Run all four monitored packages unless the user narrowed scope: env -u UV_EXCLUDE_NEWER -u UV_EXCLUDE_NEWER_PACKAGE \ uv run --locked --extra codex python -m examples.sdk_evolution_agent \ --runtime codex-agent-sdk \ + --model gpt-6.1-sol \ + --reasoning-effort max \ --refresh-preview \ --inspect-candidates \ --package claude-agent-sdk \ @@ -157,6 +160,8 @@ Once the adaptation and evidence support promotion, run the implementation pass: env -u UV_EXCLUDE_NEWER -u UV_EXCLUDE_NEWER_PACKAGE \ uv run --locked --extra codex python -m examples.sdk_evolution_agent \ --runtime codex-agent-sdk \ + --model gpt-6.1-sol \ + --reasoning-effort max \ --refresh-preview \ --inspect-candidates \ --implementation-enabled \ diff --git a/docs/sdk-evolution-agent-design.md b/docs/sdk-evolution-agent-design.md index f952913..3772dd5 100644 --- a/docs/sdk-evolution-agent-design.md +++ b/docs/sdk-evolution-agent-design.md @@ -194,9 +194,9 @@ that exact home, and mirrors the supported normal Codex login cache from ask for access tokens, or inspect unsupported credential stores. When `codex-agent-sdk` is selected for SDK update work, every AI-backed stage -should run on `gpt-5.5` with `reasoning_effort=xhigh`. This is a Codex runtime +should run on `gpt-6.1-sol` with `reasoning_effort=max`. This is a Codex runtime policy, not a portable metadata field: Claude and Antigravity runs should not -receive a `gpt-5.5` model override. +receive a `gpt-6.1-sol` model override. Package filters narrow evidence collection for debugging, but normal evolution runs should inspect all tracked packages: diff --git a/docs/sdk-evolution-agent.md b/docs/sdk-evolution-agent.md index 055159b..c788536 100644 --- a/docs/sdk-evolution-agent.md +++ b/docs/sdk-evolution-agent.md @@ -66,15 +66,15 @@ env -u UV_EXCLUDE_NEWER \ uv run --locked --extra codex python -m examples.sdk_evolution_agent.auth ensure-codex ``` -Codex-backed SDK evolution runs default to `gpt-5.5` with -`reasoning_effort=xhigh` for the AI stages that analyze direction, decide the +Codex-backed SDK evolution runs default to `gpt-6.1-sol` with +`reasoning_effort=max` for the AI stages that analyze direction, decide the update plan, and review the result. Dependency application and verification are deterministic. This model policy is applied only to `codex-agent-sdk`; Claude and Antigravity runs keep their -provider-native model selection because `gpt-5.5` is not a valid model override +provider-native model selection because `gpt-6.1-sol` is not a valid model override for those adapters. Use `--model` and `--reasoning-effort` to select the requested model explicitly -for every AI stage. For example, use `--model gpt-6-astra --reasoning-effort xhigh`. +for every AI stage. For example, use `--model gpt-6.1-sol --reasoning-effort max`. These options populate the first-class `AgentTask` fields and are recorded in `config.json`; unsupported provider values remain provider errors. diff --git a/examples/sdk_evolution_agent/stages.py b/examples/sdk_evolution_agent/stages.py index 3675ed5..5aeb528 100644 --- a/examples/sdk_evolution_agent/stages.py +++ b/examples/sdk_evolution_agent/stages.py @@ -57,8 +57,8 @@ class StageExecutionError(RuntimeError): SDK_EVOLUTION_CODEX_HOME = Path("~/.codex_agent_runtime_sdk").expanduser() -SDK_EVOLUTION_CODEX_MODEL = "gpt-5.5" -SDK_EVOLUTION_CODEX_REASONING_EFFORT = "xhigh" +SDK_EVOLUTION_CODEX_MODEL = "gpt-6.1-sol" +SDK_EVOLUTION_CODEX_REASONING_EFFORT = "max" class FixtureEvolutionRuntime: diff --git a/tests/test_sdk_evolution_agent.py b/tests/test_sdk_evolution_agent.py index fbcea4e..edd354a 100644 --- a/tests/test_sdk_evolution_agent.py +++ b/tests/test_sdk_evolution_agent.py @@ -3230,7 +3230,7 @@ async def test_stage_execution_uses_agent_task_runtime_primitives(tmp_path: Path @pytest.mark.asyncio -async def test_codex_stage_execution_uses_gpt_55_xhigh_thinking(tmp_path: Path) -> None: +async def test_codex_stage_execution_uses_gpt_61_sol_max_thinking(tmp_path: Path) -> None: runtime = RecordingRuntime(kind=AgentRuntimeKind.CODEX_AGENT_SDK) context = RunContext( run_id="run-1", @@ -3251,8 +3251,10 @@ async def test_codex_stage_execution_uses_gpt_55_xhigh_thinking(tmp_path: Path) ) assert runtime.task is not None - assert runtime.task.metadata["model"] == SDK_EVOLUTION_CODEX_MODEL - assert runtime.task.metadata["reasoning_effort"] == SDK_EVOLUTION_CODEX_REASONING_EFFORT + assert runtime.task.metadata["model"] == SDK_EVOLUTION_CODEX_MODEL == "gpt-6.1-sol" + assert ( + runtime.task.metadata["reasoning_effort"] == SDK_EVOLUTION_CODEX_REASONING_EFFORT == "max" + ) @pytest.mark.asyncio @@ -4090,16 +4092,16 @@ def test_parse_args_explicit_analysis_driver() -> None: "--runtime", "codex-agent-sdk", "--model", - "gpt-6-astra", + "gpt-6.1-sol", "--reasoning-effort", - "xhigh", + "max", "--codex-bin", "/installed/codex", ] ) assert (options.model, options.reasoning_effort, options.codex_bin) == ( - "gpt-6-astra", - "xhigh", + "gpt-6.1-sol", + "max", "/installed/codex", ) with pytest.raises(SystemExit):