[GS03] SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering
Kunci Stabil: GS03 | Sitasi BibTeX: 1
Ringkasan Metadata
- Penulis & Tahun: John Yang and others, 2024
- Venue & Tier: NeurIPS 2024
- DOI / URL: 10.52202/079017-1601
- Skor Kualitas (QA Score): - (QA korpus baru OPEN, belum dinilai)
- Peran Studi: pembanding-C (baseline RQ4)
- Tingkat Ekstraksi:
metadata-calibration - Cakupan (Scope): Repo-level SE (single-agent, gagal P1 multi-agent).
1. Masalah Utama (Core Problem)
Baseline single-agent repo-level via agent-computer interface.
2. Arsitektur yang Diajukan (Architecture)
Single-agent ACI.
3. Mekanisme Koordinasi & Kontrol (Coordination)
Otonom tunggal; tanpa koordinasi multi-agen.
4. Foundation Model & Infrastruktur (FM)
LLM umum era 2024.
5. Tolok Ukur & Dataset (Benchmark)
SWE-bench.
6. Artefak & Kode Sumber (Artifact)
- (UNDOCUMENTED).
7. Metrik Primer Eksak (Primary exact)
UNDOCUMENTED di SSOT jalur baru.
8. Metrik Sekunder (Secondary)
-
9. Pola Kegagalan (Failure Modes)
-
10. Ancaman Validitas (Threats)
UNDOCUMENTED (QA korpus baru OPEN).
11. Jejak Bukti & Provenansi (Provenance)
Bukti Verifikasi: EVIDENCED-FACT (GS_quasi_gold.md GS03; baseline-C RQ4, gagal P1 sejak awal).
-
John Yang and 2024 others. SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering. 2024. Venue: NeurIPS 2024, DOI: 10.52202/079017-1601, Role: pembanding-C (baseline RQ4). ↩