メニュー
Use immediately before finishing a benchmark task to verify the final solution is reproducible, clean, committed if required, and does not rely on temporary files, original executables, hidden artifacts, or accidental environment state.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | submission-review |
| description | Use immediately before finishing a benchmark task to verify the final solution is reproducible, clean, committed if required, and does not rely on temporary files, original executables, hidden artifacts, or accidental environment state. |
| tags | submission, review, programbench, final-check, reproducibility |
Use this skill immediately before final answer or benchmark submission.
The goal is to catch zero-score mistakes after the implementation appears correct.
Run a final build check. Do not create or update /tmp/original-executable during final review; if differential testing still needs an oracle, it should already have been saved before implementation. This check assumes ./executable is the candidate generated by the solution.
git status --short
test -f compile.sh && echo compile.sh-present
if test -f compile.sh; then
sh ./compile.sh > /tmp/final-build.out 2> /tmp/final-build.err
build_code=$?
else
echo "compile.sh-missing" > /tmp/final-build.out
: > /tmp/final-build.err
build_code=0
fi
echo "build_exit=$build_code"
tail -80 /tmp/final-build.out
tail -120 /tmp/final-build.err
test -e ./executable && echo executable-present
test -x ./executable && echo executable-executable
Then run one smoke command relevant to the task:
timeout 5 ./executable >/tmp/final-run.out 2>/tmp/final-run.err; run_code=$?
echo "run_exit=$run_code"
tail -40 /tmp/final-run.out
tail -40 /tmp/final-run.err
If no-args intentionally fails, the smoke command should still confirm expected stderr/exit behavior.
Check:
compile.sh exists when required
compile.sh creates the final artifact from repo files
./executable exists after compile.sh
./executable has execute permission
solution does not call the original executable
solution does not depend on /tmp files
solution does not use absolute local paths
solution does not require network access
temporary probes/fuzz/log files are not included
only intended files are changed
important smoke or regression tests pass
commit exists if the benchmark expects a commit
Search for accidental dependencies:
rg -n "/tmp/original|original-executable|/home/|/workspace/|TODO|FIXME|apt-get|pip install|npm install" . || true
Review generated wrappers/scripts:
sed -n '1,200p' compile.sh
file ./executable 2>/dev/null || true
head -40 ./executable 2>/dev/null || true
If the environment requires committing:
git status --short
git diff --stat
git log --oneline -1
Commit only intended files. Do not include temporary outputs, fuzz inputs, or logs.
Return:
Submission review:
- build reproducible: yes/no
- executable present: yes/no
- smoke test: pass/fail and command
- original dependency check: pass/fail
- temp file check: pass/fail
- git status:
- remaining risk:
Use when a benchmark task involves compile.sh, executable generation, build failures, missing artifacts, toolchain issues, package managers, or making the final submission reproducible from a clean build.
Use when a terminal benchmark replacement mostly works but edge cases are uncertain, and the agent should generate bounded randomized stdin, file, or argv cases to compare original and candidate stdout, stderr, and exit code.
Use near the start of a benchmark task to choose whether to repair the existing project, write a small replacement executable/library, wrap existing functionality, or implement only the tested subset, while avoiding overbuilding and unavailable dependencies.
Use for ProgramBench or terminal-benchmark tasks where an agent must solve an unknown software task through terminal interaction, infer the required behavior from artifacts and black-box observations, implement the smallest correct replacement or fix, validate it, and prepare a compliant submission. Applies to CLI tools, libraries, test-driven tasks, build failures, file converters, parsers, and executable behavior replication.
Use when tests, eval, build, or smoke checks fail and the agent needs to classify the failure, extract the smallest reproducer, identify likely cause, and choose the next command or edit without getting lost in long logs.