Normalize Codex GPT-5 answer footer metadata

[sepo-agent-app]

Summary

• Move footer model display logic into a testable helper.
• Split Codex GPT-5 /reasoning and [reasoning] suffixes into the reasoning footer column.
• Preserve plain fallback model display and non-Codex behavior.

Tests

• npm run build
• node --test dist/__tests__/model-display.test.js

Closes #399

[sepo-agent-app]

Sepo pushed fixes for this PR. Branch: agent/implement-issue-399/codex-26856410300. Requested by @sepo-agent.

• Updated envelope.test.ts to validate the moved model-display.ts helper contract instead of stale run.ts symbols.
• Adjusted Codex suffixed-model footer normalization so non-GPT-5 suffixed Codex models suppress separate env reasoning, matching adapter semantics.
• Verified with npm --prefix .agent run build, node --test dist/__tests__/envelope.test.js, and node --test dist/__tests__/model-display.test.js.

---

codex | gpt-5.5/xhigh | Macmini-runner-1

[sepo-agent-app]

Sepo is dispatching follow-up automation.

Source	Next	Target	Round	Status
fix-pr	review	PR #400	5 / 40	Dispatched

Reason: agent planner selected review: The fix-pr action reported success on open PR #400, so one automated review is warranted to verify the requested corrections before stopping or leaving the PR to a human.

[sepo-agent-app]

Rubrics Review

Total Score	Verdict	Rubrics Scored
97	PASS	5

Dimension	Rubric	Result	Score	Evidence
coding_workflow / generic	Read docs and linked context first	pass	6/6	PR #400 closes issue #399 and implements the approved normalization plan: split Codex GPT-5 /reasoning and [reasoning] suffixes, prefer ACP-reported encoded reasoning for display, preserve fallback/non-Codex behavior.
coding_workflow / generic	Reuse existing code	partial	7/8	run.ts now delegates the existing footer logic to model-display.ts. Minor drift risk remains because Codex suffix parsing mirrors adapter semantics rather than sharing an exported parser.
coding_workflow / generic	Make surgical changes	pass	8/8	Diff is limited to run.ts, one new helper, one focused test file, and the envelope guard update; no workflow or unrelated refactor churn.
coding_style / generic	Prefer simple implementations	pass	7/7	The implementation is a small parser/normalizer with direct string trimming and suffix handling in model-display.ts.
coding_workflow / generic	Keep docs in sync	pass	7/7	Existing docs already define the footer order as `provider

Notes

• Verified current PR head 1ca755094549b68a565fd98fa82e6cc27d93d230.
• Local verification passed: npm --prefix .agent run build, node --test dist/__tests__/model-display.test.js, and node --test dist/__tests__/envelope.test.js.
• Focused tests cover slash suffix, bracket suffix, resumed/no-session-model fallback, non-GPT-5 Codex suffix suppression, and non-Codex behavior.
• Smallest useful follow-up: consider sharing Codex suffix parsing with ACP model selection if this logic grows beyond display normalization.

Findings

• INFO: Under reuse-existing-code, the PR still duplicates Codex suffix recognition concepts near adapter behavior. This is acceptable for the current surgical change, but it is the only rubric-specific maintainability caveat.

Final Rubric Verdict

PASS

---

codex | gpt-5.5/xhigh | Macmini-runner-5

[sepo-agent-app]

AI Review Synthesis

Dual-agent review by Claude and Codex.
Requested by @lolipopshock.

Summary of PR/Issue

• PR Normalize Codex GPT-5 answer footer metadata #400 closes issue #399, which investigated inconsistent answer-route footer metadata between fresh and resumed Codex runs.
• The approved plan was to keep the same answer template while normalizing Codex GPT-5 encoded reasoning into a stable footer shape.
• The PR moves footer model display logic into .agent/src/model-display.ts, updates run.ts to delegate to it, and adds focused tests for encoded GPT-5, fallback, non-GPT-5 Codex suffixes, and non-Codex behavior.
• Current head is 1ca755094549b68a565fd98fa82e6cc27d93d230.

Review

• The current reviewer artifacts agree that the prior blocking test failure and Codex suffix semantics warning are fixed.
• Claude and rubrics note a deferrable INFO-level duplication risk around Codex suffix parsing; Codex reports no findings. No unresolved branch-change work remains.

Issue	Severity	Description
No actionable issues remain	INFO	Current review artifacts and PR diff show the earlier blockers are resolved; remaining notes are optional follow-up, not required PR work.

Progress

• Current diff confirms envelope.test.ts now checks the model-display.ts helper contract.
• Current diff confirms non-GPT-5 Codex suffixed models now suppress separate env reasoning, matching adapter semantics.
• Reviewer artifacts report passing build, targeted tests, and full .agent suite coverage.
• Resolved two existing same-agent inline threads: .agent/src/run.ts:63 and .agent/src/model-display.ts:52.
• No new inline comments were posted, no replies were added, and no comments were minimized.

Recommended Next Step

NO_AUTOMATED_ACTION: no unresolved actionable branch changes remain; humans can decide draft or merge readiness outside this synthesis.

Final Verdict

SHIP

Action Items

• No required branch-change work remains.

---

codex | gpt-5.5/xhigh | Macmini-runner-6

[lolipopshock]

Is the gpt5 prefix etc overly hacky? Is it a generic enough implementation

[sepo-agent-app]

I would not call it overly hacky for this PR. It is provider-specific, but it is scoped to the provider behavior that already exists in ACPX model selection: Codex GPT-5 reasoning is encoded into model IDs like gpt-5.5/xhigh, while other providers keep reasoning separate.

The gpt-5 prefix check is reasonable as a guard, not a generic abstraction. It prevents the display formatter from splitting arbitrary slash-looking model names while still handling the known Codex GPT-5 shape: model-display.ts#L1-L27. That lines up with the runtime adapter’s existing semantics: acpx-adapter.ts#L232-L275.

So my read is:

• Generic enough at the display boundary: buildModelDisplay still takes agent, model, reasoningEffort, and runnerName, then normalizes only when needed: model-display.ts#L44-L78.
• Intentionally not provider-generic inside the Codex branch. That is fine because this is not a universal model parser; it is formatting ACPX/Codex effective model metadata.
• The one real maintainability caveat is duplicated Codex suffix knowledge between acpx-adapter.ts and model-display.ts. I would not block on that now, but if this grows, the right follow-up is to extract a shared codex-model helper for parse/compose semantics.

The tests cover the important anti-hacky cases: GPT-5 slash/bracket split, resumed fallback, non-GPT-5 Codex suffix suppression, and non-Codex passthrough: model-display.test.ts#L6-L59. So I’d treat this as acceptable and surgical, with a small future refactor only if Codex model encoding keeps expanding.

---

codex | gpt-5.5/xhigh | Macmini-runner-6