[ { "Paper_ID": "GS02", "BID": "GS02", "CiteKey": "qian2024chatdev", "Title": "ChatDev: Communicative Agents for Software Development", "Authors_Year": "Chen Qian and others, 2024", "Venue_Tier": "ACL 2024 Long", "DOI_URL": "10.18653/v1/2024.acl-long.810", "Core_Problem": "Pengembangan perangkat lunak kolaboratif oleh agen komunikatif berbasis peran; cakupan greenfield multi-berkas skala kecil, bukan maintenance repo besar.", "Scope": "SE development scope (greenfield multi-file, non-SWE-bench)", "Proposed_Architecture": "Waterfall role-play komunikatif (CEO, CTO, CPO, Programmer, Reviewer, Tester) via chat chain.", "Coordination_Control": "Orkestrasi waterfall linier antar-peran; otonom antar-agen, tanpa supervisor dinamis dan tanpa approval gate selektif.", "Foundation_Model": "LLM umum era 2024 (laporan facet-map subagent: durasi ~7 mnt, biaya $0.18-$0.30, code completeness di SRDD — BELUM jangkar PDF).", "Benchmark_Dataset": "SRDD (aplikasi/game creation skala kecil).", "Artifact_Type": "OpenBMB/ChatDev.", "Primary_Metrics_Exact": "UNDOCUMENTED di SSOT jalur baru (QA/DEF korpus baru OPEN); landmark waterfall role-based.", "Secondary_Metrics": "Durasi/biaya generasi per rilis kecil (lihat naskah sumber).", "Failure_Modes": "Cascading error + halusinasi berantai tanpa gerbang pemutus.", "Threats_Validity": "UNDOCUMENTED (QA korpus baru OPEN).", "Provenance_Tag": "EVIDENCED-FACT (GS_quasi_gold.md GS02; CrossRef DOI resolve; Scopus-769 HIT DOI eksak via micro-fix P2 v2.2).", "QA_Score": "-", "Decision": "QGS-ELIGIBLE", "Role": "QGS (penguji recall P1+P2)", "Extraction_Level": "metadata-calibration", "txt_path": "", "year_derived": 2024, "tags_derived": [ "qgs", "role-based", "waterfall" ] }, { "Paper_ID": "GS05", "BID": "GS05", "CiteKey": "tao2024magis", "Title": "MAGIS: LLM-Based Multi-Agent Framework for GitHub Issue Resolution", "Authors_Year": "Wei Tao and others, 2024", "Venue_Tier": "NeurIPS 2024 (Vol 37)", "DOI_URL": "10.52202/079017-1647", "Core_Problem": "Resolusi GitHub issue repo-level secara otonom.", "Scope": "Repository-level GitHub issue resolution (SWE-bench 500 instances).", "Proposed_Architecture": "Hierarki 4 agen: Manager, Custodian, Developer, QA (supervisor-worker).", "Coordination_Control": "Manager membagi tugas dan mengarahkan agen lain; Custodian membatasi search space berkas; 100% otonom tanpa HITL gate.", "Foundation_Model": "GPT-4 era.", "Benchmark_Dataset": "SWE-bench (500 issue repo-level).", "Artifact_Type": "weitao92/MAGIS.", "Primary_Metrics_Exact": "Resolved rate 13.94% = 8x GPT-4 1.74% (PDF hal 1+7/Table 2 hal 7).", "Secondary_Metrics": "-", "Failure_Modes": "86% isu gagal terselesaikan tanpa batas eskalasi terukur.", "Threats_Validity": "UNDOCUMENTED (QA korpus baru OPEN).", "Provenance_Tag": "EVIDENCED-FACT (GS_quasi_gold.md GS05; Scopus-769 HIT judul eksak; arXiv 2403.17927v2 HIT).", "QA_Score": "-", "Decision": "QGS-ELIGIBLE", "Role": "QGS (penguji recall P1+P2)", "Extraction_Level": "metadata-calibration", "txt_path": "gs_pdfs/MAGIS_NeurIPS2024.pdf", "year_derived": 2024, "tags_derived": [ "qgs", "supervisor-worker", "swe-bench" ] }, { "Paper_ID": "GS06", "BID": "GS06", "CiteKey": "zhang2026sgagent", "Title": "SGAgent: Suggestion-Guided LLM-Based Multi-Agent Framework for Repository-Level Software Repair", "Authors_Year": "Zhang and others (NJU), 2026", "Venue_Tier": "ACM TOSEM (Q1) 2026", "DOI_URL": "10.1145/3818617", "Core_Problem": "Repository-level software repair.", "Scope": "SWE-bench Lite (300 instances) & VUL4J.", "Proposed_Architecture": "Localize-suggest-fix berfase (Localizer, Suggester, Fixer) + RepoGraph KG.", "Coordination_Control": "Koordinasi sekuensial suggestion-guided berbasis graf pengetahuan; otonom penuh tanpa hierarki komando dan tanpa approval gate.", "Foundation_Model": "Claude-3.5 / Claude-4 (evaluasi).", "Benchmark_Dataset": "SWE-bench Lite + VUL4J.", "Artifact_Type": "NJU-Seq/SGAgent.", "Primary_Metrics_Exact": "Lite 51.3% (Claude-3.5) s.d. 60.7% (Claude-4) + VUL4J 48.0% (abstrak resmi CrossRef; PDF paywalled ACM, BELUM jangkar halaman).", "Secondary_Metrics": "-", "Failure_Modes": "Gagal bila saran awal menyesatkan fixer tanpa verifikasi eksternal.", "Threats_Validity": "0 sitasi (umur 3,5 bln); PDF paywalled ACM.", "Provenance_Tag": "EVIDENCED-FACT (GS_quasi_gold.md GS06; CrossRef resolve; arXiv 2602.23647v2 HIT; Scopus MISS = lag indeks TOSEM 2026).", "QA_Score": "-", "Decision": "QGS-ELIGIBLE", "Role": "QGS (penguji recall P1+P2)", "Extraction_Level": "metadata-calibration", "txt_path": "", "year_derived": 2026, "tags_derived": [ "qgs", "knowledge-graph", "swe-bench" ] }, { "Paper_ID": "GS07", "BID": "GS07", "CiteKey": "islam2024mapcoder", "Title": "MapCoder: Multi-Agent Code Generation for Competitive Problem Solving", "Authors_Year": "Md Ashraful Islam and others (BUET & Rochester), 2024", "Venue_Tier": "ACL 2024 Long", "DOI_URL": "10.18653/v1/2024.acl-long.269", "Core_Problem": "Generasi kode problem-solving kompetitif via multi-agen.", "Scope": "Competitive problem solving (HumanEval, MBPP, APPS, CodeContests) — function-level, bukan repo.", "Proposed_Architecture": "4 agen (Retrieval, Planning, Coding, Debugging); siklus plan-code-debug retrieval-augmented.", "Coordination_Control": "Siklik retrieval-planning; otonom penuh tanpa hierarki formal dan tanpa gate.", "Foundation_Model": "LLM umum era 2024.", "Benchmark_Dataset": "HumanEval, MBPP, APPS, CodeContests.", "Artifact_Type": "ideis/MapCoder.", "Primary_Metrics_Exact": "UNDOCUMENTED di SSOT jalur baru (QA/DEF korpus baru OPEN).", "Secondary_Metrics": "-", "Failure_Modes": "Generalisasi ke repo skala besar belum terbukti.", "Threats_Validity": "UNDOCUMENTED (QA korpus baru OPEN).", "Provenance_Tag": "EVIDENCED-FACT (GS_quasi_gold.md GS07; Scopus-769 HIT DOI eksak; arXiv 2405.11403v1 HIT).", "QA_Score": "-", "Decision": "QGS-ELIGIBLE", "Role": "QGS (penguji recall P1+P2)", "Extraction_Level": "metadata-calibration", "txt_path": "", "year_derived": 2024, "tags_derived": [ "qgs", "retrieval-planning" ] }, { "Paper_ID": "GS10", "BID": "GS10", "CiteKey": "lee2025unidebugger", "Title": "UniDebugger: Hierarchical Multi-Agent Framework for Unified Software Debugging", "Authors_Year": "Cheryl Lee and others (CUHK & UIUC), 2025", "Venue_Tier": "EMNLP 2025 Main", "DOI_URL": "10.18653/v1/2025.emnlp-main.921", "Core_Problem": "Unified software debugging multi-berkas.", "Scope": "APR multi-berkas Defects4J v1.2 & v2.0 (bukan open-ended issue resolving).", "Proposed_Architecture": "Hierarki adaptif 3 level (L1 simple, L2 single-file, L3 cross-file MAS).", "Coordination_Control": "Eskalasi penugasan L1-L2-L3 sesuai kesulitan bug; 100% otonom tanpa gate manusia.", "Foundation_Model": "LLM umum era 2025.", "Benchmark_Dataset": "Defects4J (197 correct / 286 plausible).", "Artifact_Type": "Cheryl-Lee/UniDebugger.", "Primary_Metrics_Exact": "197 correct / 286 plausible = +25.48% vs ChatRepair + 42 unik (Figure 5 hal 7).", "Secondary_Metrics": "-", "Failure_Modes": "Patch plausible-but-incorrect lolos tanpa verifikasi manusia.", "Threats_Validity": "Sitasi 2 (prosiding baru).", "Provenance_Tag": "EVIDENCED-FACT (GS_quasi_gold.md GS10; Scopus-769 HIT DOI eksak; arXiv 2404.17153v3 HIT).", "QA_Score": "-", "Decision": "QGS-ELIGIBLE", "Role": "QGS (penguji recall P1+P2)", "Extraction_Level": "metadata-calibration", "txt_path": "gs_pdfs/UniDebugger_EMNLP2025.pdf", "year_derived": 2025, "tags_derived": [ "qgs", "hierarchical", "debugging" ] }, { "Paper_ID": "GS11", "BID": "GS11", "CiteKey": "kumar2026balancing", "Title": "Balancing autonomy and oversight in reliable agentic AI through adaptive human interaction architectures", "Authors_Year": "Madhvesh Kumar, Deepika Singh, 2026", "Venue_Tier": "Discover Artificial Intelligence (Springer) Vol 6", "DOI_URL": "10.1007/s44163-026-01373-2", "Core_Problem": "Adaptive human interaction: kapan agen jalan mandiri vs eskalasi ke operator.", "Scope": "Enterprise code generation & workflow automation (5.000 tugas terstruktur).", "Proposed_Architecture": "Supervisor-Worker multi-agent + Dynamic Intervention Framework (Contextual Confidence Score).", "Coordination_Control": "Supervisor menilai tiap langkah worker; skor > ambang atas auto-approve, zona tengah post-hoc audit flag, di bawah ambang bawah halt-eskalasi ke operator manusia.", "Foundation_Model": "LLM umum + DPO feedback.", "Benchmark_Dataset": "5.000 tugas terstruktur (codegen + finance/services).", "Artifact_Type": "- (UNDOCUMENTED).", "Primary_Metrics_Exact": "Intervensi 21.0%->11.5% + success 98.2% (abstrak Scopus; PDF paywalled, BELUM jangkar halaman).", "Secondary_Metrics": "Hanya 14.5% langkah keputusan butuh intervensi.", "Failure_Modes": "- (UNDOCUMENTED).", "Threats_Validity": "0 sitasi (umur 3,5 bln); jurnal AI umum, bukan flagship SE.", "Provenance_Tag": "EVIDENCED-FACT (CrossRef resolve 2026-09-12; disjoint-47; Scopus-769 HIT EID 2-s2.0-105045578545).", "QA_Score": "-", "Decision": "QGS-ELIGIBLE", "Role": "QGS (penutup lubang I-gate)", "Extraction_Level": "metadata-calibration", "txt_path": "", "year_derived": 2026, "tags_derived": [ "qgs", "selective-gate", "supervisor-worker" ] }, { "Paper_ID": "GS01", "BID": "GS01", "CiteKey": "li2023camel", "Title": "CAMEL: Communicative Agents for \"Mind\" Exploration of Large Language Model Society", "Authors_Year": "Guohao Li and others, 2023", "Venue_Tier": "NeurIPS 2023", "DOI_URL": "10.52202/075280-2264", "Core_Problem": "Role-play kolaboratif multi-agen komunikatif.", "Scope": "Fondasi umum (di luar SE-spesifik).", "Proposed_Architecture": "Communicative agents role-play.", "Coordination_Control": "Koordinasi komunikatif antar-peran.", "Foundation_Model": "LLM umum era 2023.", "Benchmark_Dataset": "- (fondasi umum).", "Artifact_Type": "- (UNDOCUMENTED).", "Primary_Metrics_Exact": "UNDOCUMENTED di SSOT jalur baru.", "Secondary_Metrics": "-", "Failure_Modes": "-", "Threats_Validity": "UNDOCUMENTED (QA korpus baru OPEN).", "Provenance_Tag": "EVIDENCED-FACT (GS_quasi_gold.md GS01; pembanding-C, bukan denominator recall).", "QA_Score": "-", "Decision": "COMPARISON-C", "Role": "pembanding-C (fondasi Bab 2)", "Extraction_Level": "metadata-calibration", "txt_path": "", "year_derived": 2023, "tags_derived": [ "pembanding-c", "fondasi" ] }, { "Paper_ID": "GS03", "BID": "GS03", "CiteKey": "yang2024sweagent", "Title": "SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering", "Authors_Year": "John Yang and others, 2024", "Venue_Tier": "NeurIPS 2024", "DOI_URL": "10.52202/079017-1601", "Core_Problem": "Baseline single-agent repo-level via agent-computer interface.", "Scope": "Repo-level SE (single-agent, gagal P1 multi-agent).", "Proposed_Architecture": "Single-agent ACI.", "Coordination_Control": "Otonom tunggal; tanpa koordinasi multi-agen.", "Foundation_Model": "LLM umum era 2024.", "Benchmark_Dataset": "SWE-bench.", "Artifact_Type": "- (UNDOCUMENTED).", "Primary_Metrics_Exact": "UNDOCUMENTED di SSOT jalur baru.", "Secondary_Metrics": "-", "Failure_Modes": "-", "Threats_Validity": "UNDOCUMENTED (QA korpus baru OPEN).", "Provenance_Tag": "EVIDENCED-FACT (GS_quasi_gold.md GS03; baseline-C RQ4, gagal P1 sejak awal).", "QA_Score": "-", "Decision": "COMPARISON-C", "Role": "pembanding-C (baseline RQ4)", "Extraction_Level": "metadata-calibration", "txt_path": "", "year_derived": 2024, "tags_derived": [ "pembanding-c", "baseline", "single-agent" ] }, { "Paper_ID": "GS04", "BID": "GS04", "CiteKey": "zhang2024autocoderover", "Title": "AutoCodeRover: Autonomous Program Improvement", "Authors_Year": "Yuntong Zhang and others, 2024", "Venue_Tier": "ISSTA 2024", "DOI_URL": "10.1145/3650212.3680384", "Core_Problem": "Baseline autonomous program repair (AST-search, single-agent).", "Scope": "Repo-level APR (single-agent, gagal P1 multi-agent).", "Proposed_Architecture": "Autonomous AST-search.", "Coordination_Control": "Otonom tunggal; tanpa koordinasi multi-agen.", "Foundation_Model": "LLM umum era 2024.", "Benchmark_Dataset": "SWE-bench (repo-level APR).", "Artifact_Type": "- (UNDOCUMENTED).", "Primary_Metrics_Exact": "UNDOCUMENTED di SSOT jalur baru.", "Secondary_Metrics": "-", "Failure_Modes": "-", "Threats_Validity": "UNDOCUMENTED (QA korpus baru OPEN).", "Provenance_Tag": "EVIDENCED-FACT (GS_quasi_gold.md GS04; baseline-C RQ4, gagal P1 sejak awal).", "QA_Score": "-", "Decision": "COMPARISON-C", "Role": "pembanding-C (baseline RQ4)", "Extraction_Level": "metadata-calibration", "txt_path": "", "year_derived": 2024, "tags_derived": [ "pembanding-c", "baseline", "single-agent" ] }, { "Paper_ID": "GS08", "BID": "GS08", "CiteKey": "shinn2023reflexion", "Title": "Reflexion: Language Agents with Verbal Reinforcement Learning", "Authors_Year": "Noah Shinn and others, 2023", "Venue_Tier": "NeurIPS 2023", "DOI_URL": "10.52202/075280-0377", "Core_Problem": "Fondasi feedback-loop / self-reflection agen bahasa.", "Scope": "Fondasi umum (di luar SE-spesifik).", "Proposed_Architecture": "Verbal reinforcement learning (self-reflection loop).", "Coordination_Control": "Loop refleksi verbal mandiri.", "Foundation_Model": "LLM umum era 2023.", "Benchmark_Dataset": "- (fondasi umum).", "Artifact_Type": "- (UNDOCUMENTED).", "Primary_Metrics_Exact": "UNDOCUMENTED di SSOT jalur baru.", "Secondary_Metrics": "-", "Failure_Modes": "-", "Threats_Validity": "UNDOCUMENTED (QA korpus baru OPEN).", "Provenance_Tag": "EVIDENCED-FACT (GS_quasi_gold.md GS08; pembanding-C, bukan denominator recall).", "QA_Score": "-", "Decision": "COMPARISON-C", "Role": "pembanding-C (fondasi Bab 2)", "Extraction_Level": "metadata-calibration", "txt_path": "", "year_derived": 2023, "tags_derived": [ "pembanding-c", "fondasi" ] }, { "Paper_ID": "GS09", "BID": "GS09", "CiteKey": "fakhoury2024ticoder", "Title": "TiCoder: Test-Driven Interactive Code Generation", "Authors_Year": "Sarah Fakhoury and others, 2024", "Venue_Tier": "IEEE TSE 2024 Vol 50 Issue 9", "DOI_URL": "10.1109/tse.2024.3428972", "Core_Problem": "Selective interaction gate (test-driven) pada code generation single-interactive.", "Scope": "Codegen test-driven + user study (gagal P1 multi-agent by-design).", "Proposed_Architecture": "Single-interactive agent via LLM + test-driven gate (m=1..5).", "Coordination_Control": "Gerbang interaksi selektif test-driven; bukan orkestrasi multi-agen.", "Foundation_Model": "LLM umum era 2024.", "Benchmark_Dataset": "Codegen test-driven + user study 15 devs.", "Artifact_Type": "- (UNDOCUMENTED).", "Primary_Metrics_Exact": "+45.97% pass@1 + user study 15 devs (TABLE IV preprint v2; BELUM jangkar halaman published).", "Secondary_Metrics": "-", "Failure_Modes": "-", "Threats_Validity": "Single-interactive (P1 via LLM, bukan multi-agent strict).", "Provenance_Tag": "EVIDENCED-FACT (GS_quasi_gold.md GS09; pembanding-C selective-gate Bab 2/3).", "QA_Score": "-", "Decision": "COMPARISON-C", "Role": "pembanding-C (selective-gate Bab 2/3)", "Extraction_Level": "metadata-calibration", "txt_path": "gs_pdfs/TiCoder_TSE2024_preprint.pdf", "year_derived": 2024, "tags_derived": [ "pembanding-c", "selective-gate" ] } ]