Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 7 additions & 2 deletions .codex/skills/agent-runtime-kit-upgrade/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -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`
Expand All @@ -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
Expand All @@ -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 \
Expand Down Expand Up @@ -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 \
Expand Down
4 changes: 2 additions & 2 deletions docs/sdk-evolution-agent-design.md
Original file line number Diff line number Diff line change
Expand Up @@ -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:
Expand Down
8 changes: 4 additions & 4 deletions docs/sdk-evolution-agent.md
Original file line number Diff line number Diff line change
Expand Up @@ -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.

Expand Down
4 changes: 2 additions & 2 deletions examples/sdk_evolution_agent/stages.py
Original file line number Diff line number Diff line change
Expand Up @@ -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:
Expand Down
16 changes: 9 additions & 7 deletions tests/test_sdk_evolution_agent.py
Original file line number Diff line number Diff line change
Expand Up @@ -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",
Expand All @@ -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
Expand Down Expand Up @@ -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):
Expand Down
Loading