Retire Gateway-owned Codex tool schemas and scheduling where the spike proves native behavior #59

Problem

CodexHub currently owns Codex-specific tool behavior beyond protocol adaptation:

• hard-coded multi-agent schemas;
• explicit tool injection/filtering;
• transcript-derived lifecycle state;
• forced next-action and tool_choice repair;
• duplicate-spawn suppression;
• synthesized lifecycle/tool events;
• an independent scheduler and Dynamic DAG implementation.

Some behavior was introduced because third-party models did not call tool_search or did not reliably choose spawn/wait/close after results. It must not be deleted based on Codex++ or source inspection alone: Codex++ does not solve deferred discovery, search-disabled Codex does not automatically expose every Deferred tool, and #57 must distinguish visibility failures from provider/runtime/model failures.

src-python/subagent_dynamic_dag.py also contains production behavior gated by LEVEL3_DYNAMIC_DAG with hard-coded diagnostic nodes/outputs, which appears test-specific but must still be removed through an evidence-backed change.

Dependency gate

Blocked by #57 and #58. Do not implement until #57 has a maintainer GO/scoped PARTIAL decision and #58 has established the supported native-route contract.

For any behavior not explicitly covered by those decisions, preserve the current compatibility implementation. If deciding whether a behavior is necessary requires new live evidence, create a narrower spike first; do not delete it speculatively in this issue.

Scope

Only for provider/model/protocol/tool-class combinations proven safe:

• make live Codex schemas/history the source of truth;
• remove hard-coded schema replacement where native search or proven live direct exposure works;
• remove or disable-by-default Gateway-owned spawn/wait/close scheduling where Codex runtime plus the model completes the lifecycle;
• stop forcing next calls or synthesizing tool calls/events as a normal orchestration mechanism;
• preserve generic request-scoped namespace/name/call-id mapping only where required by the upstream protocol;
• move LEVEL3_DYNAMIC_DAG and hard-coded diagnostic nodes out of production code, or replace their purpose with replay fixtures;
• retain evidence-backed compatibility behind an explicit provider/model/tool-class mode with telemetry and rollback.

Do not mix native route enablement from #58 into this cleanup.

Removal evidence ledger

Before deleting or disabling each behavior, record:

• the exact function/module/feature being removed;
• the Spike complete Codex model capability exposure on third-party providers #57 trace/decision proving it unnecessary;
• the 0.1.8 Beta2: Implement runtime-derived tool compatibility for custom Responses endpoints #58 route and tool-class scope where removal applies;
• the regression fixture replacing the old protection;
• fallback behavior for unknown or unsupported combinations.

Absence of a failure in unrelated tests is not sufficient evidence.

Acceptance criteria

• Every removed behavior has an entry in the evidence ledger.
• Unknown, untested, or inconclusive combinations retain compatibility behavior.
• Live Codex schemas/history replace hard-coded copies only on approved capability classes.
• Gateway no longer forces or fabricates lifecycle calls only on approved native paths.
• Search-enabled and search-disabled behavior are not treated as equivalent.
• subagent_dynamic_dag.py no longer puts marker-driven diagnostics in production routing.
• Tools, namespaces, history, tool choice, call ids, SSE, completion notifications, and parent aggregation remain covered.
• Diagnostics distinguish discovery, provider loss, runtime behavior, and model orchestration weakness.
• Rollback to compatibility mode is documented.

Required verification

python -m pytest tests/test_codex_semantic_adapter.py tests/test_subagent_protocol.py tests/test_subagent_state.py tests/test_subagent_scheduler.py tests/test_subagent_dynamic_dag.py tests/test_routing.py tests/test_chat_completions_gateway.py
python -m pytest
python scripts/report_quality_gates.py

Related

• Spike complete Codex model capability exposure on third-party providers #57
• 0.1.8 Beta2: Implement runtime-derived tool compatibility for custom Responses endpoints #58
• Extract protocol_translation.py seam for Responses and Chat conversion #25
• Preserve Responses and Chat semantic fields and reject unsafe missing tool IDs #26
• src-python/codex_proxy.py
• src-python/subagent_protocol.py
• src-python/subagent_state.py
• src-python/subagent_scheduler.py
• src-python/subagent_dynamic_dag.py