Parent
#1710
Question
The capstone decision: which transport replaces codex app-server behind the flag?
Candidates, informed by the research tickets:
codex exec --json one-shots: fastest, leanest, weakest continuity and narration.codex mcp-server: sessioned via threadId, lightweight protocol, unknown event depth.- Own-the-loop via Responses API (
backend-api/codex + OAuth): maximal token and TTFT control, approvals become ours, but we rebuild the tool loop.
- Slimmed app-server: keep everything, attack handshake cost, context growth, and effort defaults directly.
Inputs: benchmark attribution, the three research findings, the narration parity bar, and the approval middle ground. Grill the trade-offs, then lock one, or lock a primary plus fallback.
Answer
Recorded on resolution: the chosen transport, the parity conditions it must meet, and what the spec-to-handoff covers.
Parent
#1710
Question
The capstone decision: which transport replaces
codex app-serverbehind the flag?Candidates, informed by the research tickets:
codex exec --jsonone-shots: fastest, leanest, weakest continuity and narration.codex mcp-server: sessioned via threadId, lightweight protocol, unknown event depth.backend-api/codex+ OAuth): maximal token and TTFT control, approvals become ours, but we rebuild the tool loop.Inputs: benchmark attribution, the three research findings, the narration parity bar, and the approval middle ground. Grill the trade-offs, then lock one, or lock a primary plus fallback.
Answer
Recorded on resolution: the chosen transport, the parity conditions it must meet, and what the spec-to-handoff covers.