{"slug": "mini-swe-agent", "name": "mini-swe-agent", "category": "ai-coding", "category_name": "AI coding tools", "description": "", "recommenders": 3, "commenters": 5, "warners": 0, "score": 3.041, "rising": -0.98, "first": "2025-12-01", "latest": "2026-09-18", "why": "Check out [1] a 100 LoC \"LLM in a loop with just terminal access\", it is now above last year's heavily harnessed SotA.", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 4, "neutral": 1, "compare": 1, "warn": 0}, "spark": [0, 1, 2, 0, 0, 0, 1, 0, 0, 0, 0, 0], "listed": true, "loved": [], "caveat": "", "short_quote": "Check out [1] a 100 LoC \"LLM in a loop with just terminal access\", it is now above last year's heavily harnessed SotA.", "kind": "", "icon": true, "domain": null, "github_repo": "github.com/swe-agent/mini-swe-agent", "official_url": "https://github.com/SWE-agent/mini-swe-agent", "pricing": "", "aliases": ["mini-swe-agent"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 772611264, "item_id": 48288204, "url": "https://news.ycombinator.com/item?id=48288204", "by": "gertlabs", "date": "2026-05-26", "stance": "recommend", "quote": "normalizing the harness to mini-swe-agent", "excerpt": "While this benchmark has interesting results, the \"Contamination free\" label only works for the initial release of the benchmark. It still has the same fundamental design issues of any other benchmark-- there's a single correct answer for tasks. It looks to be largely saturated upon release. What they did well: normalizing the harness to mini-swe-agent -- models should be able to generalize to different tools at this", "context": "", "use_case": "", "switched_from": "", "story": "DeepSWE: A contamination-free benchmark for long-horizon coding agents"}, {"id": 747017440, "item_id": 46688590, "url": "https://news.ycombinator.com/item?id=46688590", "by": "NitpickLawyer", "date": "2026-01-19", "stance": "recommend", "quote": "There's a ~100LoC harness called mini-swe-agent [1] that does just that.", "excerpt": "100% by better models. Since his talk models have gained more context windows (up to usable 1M), and RL (reinforcement learning) has been amazing at both picking out good traces, and taught the LLMs how to backtrack and overcome earlier wrong tokens. On top of that, RLAIF (RL with AI feedback) made earlier models better and RLVR (RL with verifiable rewards) has made them very good at both math and coding. The harness", "context": "", "use_case": "", "switched_from": "", "story": "Scaling long-running autonomous coding"}, {"id": 745976064, "item_id": 46623504, "url": "https://news.ycombinator.com/item?id=46623504", "by": "NitpickLawyer", "date": "2026-01-14", "stance": "recommend", "quote": "Check out mini-swe-agent.", "excerpt": "Not really. A 100 loc \"harness\" that is basically a llm in a loop with just a \"bash\" tool is way better today than the best agentic harness of last year. Check out mini-swe-agent.", "context": "", "use_case": "Minimal coding agent", "switched_from": "", "story": "Claude Cowork Exfiltrates Files"}, {"id": 744746048, "item_id": 46546628, "url": "https://news.ycombinator.com/item?id=46546628", "by": "aszen", "date": "2026-01-08", "stance": "recommend", "quote": "Look at https://github.com/SWE-agent/mini-swe-agent for proof", "excerpt": "I don't think so, model improvements far outweigh any harness or tooling. Look at https://github.com/SWE-agent/mini-swe-agent for proof", "context": "", "use_case": "", "switched_from": "", "story": "How to Code Claude Code in 200 Lines of Code"}, {"id": 741736688, "item_id": 46358543, "url": "https://news.ycombinator.com/item?id=46358543", "by": "NitpickLawyer", "date": "2025-12-22", "stance": "recommend", "quote": "Check out [1] a 100 LoC \"LLM in a loop with just terminal access\", it is now above last year's heavily harnessed SotA.", "excerpt": "> This is why we're seeing the bulk of gains from things like MCP and, now, \"agents\". This is objectively not true. The models have improved a ton (with data from \"tools\" and \"agentic loops\", but it's still the models that become more capable). Check out [1] a 100 LoC \"LLM in a loop with just terminal access\", it is now above last year's heavily harnessed SotA. > Gemini 3 Pro reaches 74% on SWE-bench verified with mi", "context": "", "use_case": "coding agent", "switched_from": "", "story": "Claude Code gets native LSP support"}]}], "caveats": [], "who": [], "tags": [], "use_cases": ["minimal coding agent", "coding agent"], "compared": [], "makers": [{"id": 796095888, "item_id": 49755993, "url": "https://news.ycombinator.com/item?id=49755993", "by": "lieret", "date": "2026-09-18", "stance": "recommend", "quote": "I wrote https://github.com/swe-agent/mini-swe-agent/ , which is probably the most minimal agent out there", "excerpt": "Cool study, we definitely need more principled studies on the role of harnesses. I'd also say that there aren't too many benchmarks where the more complicated harnesses consistently outperform extremely simple agents. But I'm also biased, because I wrote https://github.com/swe-agent/mini-swe-agent/ , which is probably the most minimal agent out there (it started as just 100 lines, all included), and it's used in a lo", "context": "author", "use_case": "coding agent benchmarks", "switched_from": "", "story": "An empirical study of harness design for coding agents"}, {"id": 744744848, "item_id": 46546553, "url": "https://news.ycombinator.com/item?id=46546553", "by": "ofirpress", "date": "2026-01-08", "stance": "recommend", "quote": "I think it's a great way to dive into the agent world", "excerpt": "We (the SWE-bench team) have a 100 line of code agent that is now pretty popular in both academic and industry labs: https://github.com/SWE-agent/mini-swe-agent I think it's a great way to dive into the agent world", "context": "member of the SWE-bench team", "use_case": "minimal coding agent", "switched_from": "", "story": "How to Code Claude Code in 200 Lines of Code"}], "threads": [{"url": "https://news.ycombinator.com/item?id=49753878", "title": "An empirical study of harness design for coding agents", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=46545620", "title": "How to Code Claude Code in 200 Lines of Code", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=48284939", "title": "DeepSWE: A contamination-free benchmark for long-horizon coding agents", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48141556", "title": "Parametric CAD Bench", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=46686418", "title": "Scaling long-running autonomous coding", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=46622328", "title": "Claude Cowork Exfiltrates Files", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {"recommend": 1}, "2026-01": {"recommend": 2}, "2026-02": {}, "2026-03": {}, "2026-04": {}, "2026-05": {"compare": 1, "recommend": 1}, "2026-06": {}, "2026-07": {}, "2026-08": {}, "2026-09": {"neutral": 1}, "2026-10": {}}, "quotes": [{"id": 796112544, "item_id": 49757034, "url": "https://news.ycombinator.com/item?id=49757034", "by": "czhu12", "date": "2026-09-18", "stance": "neutral", "quote": "190 lines with comments included so close enough", "excerpt": "I think they are referring to the file https://github.com/SWE-agent/mini-swe-agent/blob/main/src/minisweagent/agents/default.py 190 lines with comments included so close enough Which is the actual agent? Everything else is just glue code to connect to different LLM providers, etc", "context": "", "use_case": "", "switched_from": "", "story": "An empirical study of harness design for coding agents"}, {"id": 772611264, "item_id": 48288204, "url": "https://news.ycombinator.com/item?id=48288204", "by": "gertlabs", "date": "2026-05-26", "stance": "recommend", "quote": "normalizing the harness to mini-swe-agent", "excerpt": "While this benchmark has interesting results, the \"Contamination free\" label only works for the initial release of the benchmark. It still has the same fundamental design issues of any other benchmark-- there's a single correct answer for tasks. It looks to be largely saturated upon release. What they did well: normalizing the harness to mini-swe-agent -- models should be able to generalize to different tools at this", "context": "", "use_case": "", "switched_from": "", "story": "DeepSWE: A contamination-free benchmark for long-horizon coding agents"}, {"id": 770280835, "item_id": 48142552, "url": "https://news.ycombinator.com/item?id=48142552", "by": "mjzh", "date": "2026-05-14", "stance": "compare", "quote": "why opus 4.7 is with mini-swe-agent, not claude code", "excerpt": "interesting, per https://cadbench.ai/leaderboard, gpt5.5 is the best, not the opus 4.7, why opus 4.7 is with mini-swe-agent, not claude code.", "context": "", "use_case": "", "switched_from": "", "story": "Parametric CAD Bench"}, {"id": 747017440, "item_id": 46688590, "url": "https://news.ycombinator.com/item?id=46688590", "by": "NitpickLawyer", "date": "2026-01-19", "stance": "recommend", "quote": "There's a ~100LoC harness called mini-swe-agent [1] that does just that.", "excerpt": "100% by better models. Since his talk models have gained more context windows (up to usable 1M), and RL (reinforcement learning) has been amazing at both picking out good traces, and taught the LLMs how to backtrack and overcome earlier wrong tokens. On top of that, RLAIF (RL with AI feedback) made earlier models better and RLVR (RL with verifiable rewards) has made them very good at both math and coding. The harness", "context": "", "use_case": "", "switched_from": "", "story": "Scaling long-running autonomous coding"}, {"id": 745976064, "item_id": 46623504, "url": "https://news.ycombinator.com/item?id=46623504", "by": "NitpickLawyer", "date": "2026-01-14", "stance": "recommend", "quote": "Check out mini-swe-agent.", "excerpt": "Not really. A 100 loc \"harness\" that is basically a llm in a loop with just a \"bash\" tool is way better today than the best agentic harness of last year. Check out mini-swe-agent.", "context": "", "use_case": "Minimal coding agent", "switched_from": "", "story": "Claude Cowork Exfiltrates Files"}, {"id": 744746048, "item_id": 46546628, "url": "https://news.ycombinator.com/item?id=46546628", "by": "aszen", "date": "2026-01-08", "stance": "recommend", "quote": "Look at https://github.com/SWE-agent/mini-swe-agent for proof", "excerpt": "I don't think so, model improvements far outweigh any harness or tooling. Look at https://github.com/SWE-agent/mini-swe-agent for proof", "context": "", "use_case": "", "switched_from": "", "story": "How to Code Claude Code in 200 Lines of Code"}, {"id": 741736688, "item_id": 46358543, "url": "https://news.ycombinator.com/item?id=46358543", "by": "NitpickLawyer", "date": "2025-12-22", "stance": "recommend", "quote": "Check out [1] a 100 LoC \"LLM in a loop with just terminal access\", it is now above last year's heavily harnessed SotA.", "excerpt": "> This is why we're seeing the bulk of gains from things like MCP and, now, \"agents\". This is objectively not true. The models have improved a ton (with data from \"tools\" and \"agentic loops\", but it's still the models that become more capable). Check out [1] a 100 LoC \"LLM in a loop with just terminal access\", it is now above last year's heavily harnessed SotA. > Gemini 3 Pro reaches 74% on SWE-bench verified with mi", "context": "", "use_case": "coding agent", "switched_from": "", "story": "Claude Code gets native LSP support"}, {"id": 737900752, "item_id": 46118797, "url": "https://news.ycombinator.com/item?id=46118797", "by": "NitpickLawyer", "date": "2025-12-01", "stance": "recommend", "quote": "Check out the exercise from the swe-agent people who released a mini agent that's \"terminal in a loop\"", "excerpt": "It's 100% the models. Terminal bench is a good indication for this. There the agents get \"just a terminal tool\", and yet they still can solve lots and lots of tasks. Last year you needed lots of glue, and two years ago you needed monstrosities like langchain that worked maybe once in a blue moon, if you didn't look funny at it. Check out the exercise from the swe-agent people who released a mini agent that's \"termina", "context": "", "use_case": "minimal coding agent", "switched_from": "", "story": "AI agents find $4.6M in blockchain smart contract exploits"}]}