e2e-fuzz-original-fix-loop

name	e2e-fuzz-original-fix-loop
description	Run a deterministic fix loop for CLI original-vs-ported fuzz E2E mismatches by executing `mise run e2e-cli-original-fuzz`, collecting failing cases from `crates/seonbi-cli/tests/fixtures/fuzz_regressions.jsonl`, analyzing root causes against the original Haskell implementation in `seonbi/`, patching Rust core implementation in `crates/seonbi`, and iterating until recorded fuzz regressions pass.

E2E Fuzz Original Fix Loop

Use this skill when the user asks to repeatedly fix Rust behavior to match original seonbi based on failing e2e-cli-original-fuzz cases.

Goal

Drive a strict compatibility loop:

1. Run fuzz E2E.
2. Extract and cluster failing cases.
3. Reproduce minimal differences.
4. Find root cause in Haskell source.
5. Patch crates/seonbi (not test-only masking).
6. Re-run loop until recorded cases pass.

Preconditions

• Run from repository root.
• Ensure original binaries are configured via mise.toml env:
  • SEONBI_ORIGINAL_BIN
• Keep SEONBI_E2E_FUZZ_SEED fixed while debugging.
• Use non-destructive git workflow; do not reset unrelated changes.

Loop Procedure

1. Run fuzz E2E.

mise run e2e-cli-original-fuzz

2. If failing, inspect recorded cases.

wc -l crates/seonbi-cli/tests/fixtures/fuzz_regressions.jsonl
cat crates/seonbi-cli/tests/fixtures/fuzz_regressions.jsonl

3. Cluster mismatch patterns (root-cause hints).

jq -r '
  def norm: gsub("쑨원";"손문");
  if ((.original_stdout|norm)==.ported_stdout) then "reading-only" else "other" end
' crates/seonbi-cli/tests/fixtures/fuzz_regressions.jsonl | sort | uniq -c

4. Reproduce one representative failure with exact args and input from JSONL.

• Compare original vs ported stdout/stderr byte-equivalently.
• Reduce to the smallest input that still reproduces the mismatch.

5. Trace root cause in both implementations.

• Rust target: crates/seonbi/src/ only.
• Haskell reference: seonbi/src/Text/Seonbi/*.hs.
• Verify behavior expectation from Haskell code before patching.

6. Apply minimal Rust fix.

• Preserve existing semantics.
• Prefer localized changes in transformation logic or data loading.
• Do not "fix" by weakening e2e assertions or skipping failing cases.

7. Verify.

cargo fmt --all
cargo test -p seonbi-cli --test e2e_compare --no-run
mise run e2e-cli-original-fuzz

8. Repeat from step 1 until:

• compare_recorded_fuzz_regressions_with_original passes, and
• compare_fuzz_matrix_with_original adds no new mismatches for the chosen seed/iters.

Guardrails

• Keep compatibility as highest priority over style refactors.
• Edit crates/seonbi for behavioral fixes; avoid patching only CLI tests.
• If failures come from missing large dictionary assets (for example Git LFS pointer files), document it explicitly and add minimal compatibility fallback in core where justified.
• Keep each logical fix as a separate commit when requested.

Language Policy

• Think and reason in English while working through analysis and implementation details.
• Report progress and final outcomes to the user in Korean unless the user explicitly requests another language.

Reporting Format

Report each loop round with:

1. failing case count
2. mismatch clusters
3. root cause hypothesis and evidence (file and line references)
4. Rust changes made
5. verification commands and outcomes
6. residual risk