WT-4-codex-only: swap Claude→Codex baseline + ship 5 per-workload wrappers#84

Merged

yzs15 merged 34 commits into

paper/v3-integrationfrom

paper/v4/codex-only

Jul 8, 2026

Commits on Jul 7, 2026

WT-4-codex-only: spec (Stage 1) — 5 rounds codex review
claude
committed

Commits on Jul 8, 2026

WT-4-codex-only: plan (Stage 2) — 21 tasks TDD
claude
committed
WT-4-codex-only: plan v2 (r9-final, 9 codex-review rounds)
claude
committed
WT-4 Task 1: single_machine_codex/ scaffold (main+impl+workloads stubs)
claude
committed
WT-4 Task 2: codexPrompts (verbatim from claudePrompts) + AST anti-drift lock
claude
committed
WT-4 Task 3: impl.go dry-run + missing-bin fail-fast + LOCKED argv (byte-exact)
claude
committed
WT-4 Task 4: stderr scrub via secretscrub.Sanitize on both leak paths + forward-flag test
claude
committed
WT-4 Task 5: single_machine_codex/run.sh + build belt test
claude
committed
WT-4 Phase A code-review r1 fixes: Bearer scrub + AST type-check + require tools + drop dead import
claude
committed
WT-4 Task 6: enum swap single_machine_claude_code → single_machine_codex in matrix + schema
claude
committed
WT-4 Task 7: plan.py BASELINE_DIR swap (single_machine_codex → single_machine_codex/)
claude
committed
WT-4 Task 8: regenerate snapshot + no-secrets + path-exact tests
claude
committed
WT-4 Task 9: harness row_test + matrix_test + README literal swap
claude
committed
WT-4 Task 10: plan.py --filter-workload / --list-workloads / --results-root with allowlist + override_out_root through enumerator
claude
committed
WT-4 Task 11: run.sh --workload / --results-root / LOOM_FULLTABLE_WRAPPER_SHIM + 6 shell tests (SHIM + DISPATCH_SHIM + resume/E4 + symlink escape + duplicate + sample cap)
claude
committed
WT-4 Phase C code-review r1 fixes: exact-match workload allowlist + reject empty --workload + resume assert-11 + drop dup WORKLOADS + mktemp for case3 out
claude
committed
WT-4 Task 12: experiments/_common.sh helpers + bash unit tests (config parsers never log values)
claude
committed
WT-4 Task 13: 10 credential-bound preflight fixtures + skipped test (Task 18 flips to green)
claude
committed
WT-4 Task 13.5: .gitignore tests/eval/results/experiments/ (wrappers write here)
claude
committed
WT-4 Task 14: cross_device_code_mod.sh wrapper + wrapper-forwards + real-preflight tests
claude
committed
WT-4 Task 15: missing_parser_converter.sh wrapper
claude
committed
WT-4 Task 16: remote_data_processing.sh wrapper (+ writable-tmp + fixture preflight)
claude
committed
WT-4 Task 17: windows_only_artifact.sh wrapper + require_windows_host guard + --skip-if-not-windows
claude
committed
WT-4 Task 18: credential_bound_model.sh wrapper (route-a required, route-b unsupported)
claude
committed
WT-4 Task 19: windows-guard test table + uname-missing tolerance in _common.sh
claude
committed
WT-4 Task 19.5: env-allowlist unchanged guard (harness/env.go frozen for this PR)
claude
committed
WT-4 Phase D code-review r1 fixes: fake codex prints IMPOSSIBLE + IMPOSSIBLE scan + equals-form injection tests + safe uname in windows wrapper
claude
committed
WT-4 Task 20: experiments/README + repo-wide no-Claude-baseline grep test
claude
committed
WT-4 Task 21: handoff — route-b deferral, --reps deferral, paper_writing PR, P2/P3 disposition, 4-phase code-review outcomes
claude
committed
WT-4 Phase E code-review r1 fixes: repo-wide grep search whole multi-agent/ + README security wording precision
claude
committed
WT-4 fresh-review r1 fixes: env allowlist extractor + 3 P1 + P2/P3 polish
claude
committed
WT-4 fresh-review r2 fix: env allowlist extractor also strips Go comments (comment-} bypass) + shell coverage for --results-root ''
claude
committed
WT-4 fresh-review r2 fix: restore smoke fixture CSVs unintentionally deleted
claude
committed
WT-4 fresh-review r3 fix: env allowlist walker + extractor honor Go raw strings
claude
committed