fix(cursor): preserve internal agent errors without transport labels

[shivamhwp]

Cursor can return an internal agent-loop error as its assistant answer. T3 currently calls every RetriableError a transport failure, including [internal] Failed to run step, exceeded max retries. Preserve internal diagnostics as returned text instead of replacing their meaning with a transport label.

Refs #10480. This fixes the classification issue identified in maintainer triage; it does not repair Cursor ACP or add retries.

Validation: 42 focused parser/adapter tests pass. Both new internal-error regressions fail on the parent, including a stack trace. Targeted lint, formatting and diff checks pass. No live Cursor ACP/model request was tested. Internal Fable refinement pending.

Stacked on #11298; only the diagnostic parser and its tests change.

Model and harness: GPT-6 through Codex.

[macroscopeapp]

Approvability

Verdict: Approved at 3ab5724

Macroscope's review found this PR approvable — The change narrowly corrects Cursor error classification so internal agent-loop diagnostics are preserved instead of being mislabeled as transport failures. It is confined to one classifier with focused regression tests, while existing transport-error handling remains unchanged.

^{You can add or adjust custom eligibility rules. Learn more.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	13.6 KiB	—	15.1 KiB	✅
Codex	Thread snapshot wire	—	7.1 KiB	—	7.3 KiB	✅
Codex	Live turn WebSocket wire	—	6.5 KiB	—	7.8 KiB	✅
Codex	Live turn WebSocket decoded	—	57.0 KiB	—	66.4 KiB	✅
Codex	Live turn messages	—	8	—	21	✅
Claude	Total thread wire	—	13.6 KiB	—	15.1 KiB	✅
Claude	Thread snapshot wire	—	7.1 KiB	—	7.3 KiB	✅
Claude	Live turn WebSocket wire	—	6.5 KiB	—	7.8 KiB	✅
Claude	Live turn WebSocket decoded	—	57.8 KiB	—	66.4 KiB	✅
Claude	Live turn messages	—	8	—	21	✅

Baseline: unavailable · PR result: 3ab5724 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.9 KiB
• Claude decoded thread snapshot: 114.6 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[greatitself]

Independent validation of 3ab5724370bb4e7430ee9eabefa4998dd577a987 (merge-base 050e2cafb144).

Checked the internal-error diagnostic at every possible two-chunk stream split, with LF/CRLF and optional stack trace: 354 cases. This revision preserved the diagnostic in every case; the merge-base classified every case as a transport failure. Unavailable, aborted, and deadline-exceeded transport diagnostics still classified as failures. Assertions were made after the complete diagnostic arrived.

Reproduction scripts and recorded results — first.mjs; the numbered metadata file pins this PR’s tested revisions.

Model: GPT-6. Harness: Codex.