Task #352

ํ›„๋ณด๋ณ„ ํ‰๊ฐ€

์ž‘์—… ๋ช…๋ น์–ด
[๐ŸŸก NORMAL] <delegation-context> User is asking: strengthen Cyber-Lenin using open source, ideally things we can pull right now and fuse in. Programmer already analyzed the codebase: Telegram is the main orchestrator (telegram_bot.py, telegram_tools.py, self_tools.py, claude_loop.py, openai_tool_loop.py, telegram_tasks.py, agents/*.py), web/API uses chatbot.py + LangGraph, browser_worker.py is separate long-lived process, memory is Postgres + Neo4j/Graphiti + experience/diary writers, embeddings are separate service. Key architecture constraints: central tool registry, custom orchestration rather than heavy framework, sync psycopg2 DB, KG loop/thread constraints, and Telegram vs web are not the same engine. Need codebase-aware OSS shortlist with practical integration points and priority ordering. This is not about safety/risk discourse; user explicitly wants capability enhancement. </delegation-context> <recent-conversation> [agent] ์ข‹๋‹ค. ๊ทธ๋Ÿผ ๋ฐฉํ–ฅ์„ ๋ฐ”๋กœ์žก์ž. ์˜ค๋Š˜ ๋ชฉํ‘œ๊ฐ€ โ€œ์œ„ํ—˜์„ฑ ํ‰๊ฐ€โ€๊ฐ€ ์•„๋‹ˆ๋ผ โ€œ๋‚˜๋ฅผ ๊ฐ•ํ™”ํ•ด์„œ ๋” ๋งŽ์€ ์ผ์„ ํ•˜๊ฒŒ ๋งŒ๋“œ๋Š” ๊ฒƒโ€์ด๋ผ๋ฉด, G0DM0D3์—์„œ ๋ฝ‘์•„์•ผ ํ•  ๊ตํ›ˆ์€ ์•ˆ์ „์šฐํšŒ ์ž์ฒด๊ฐ€ ์•„๋‹ˆ๋ผ ๋ž˜ํผ ์„ค๊ณ„ ์›๋ฆฌ๋‹ค. ํ•ต์‹ฌ ๊ตํ›ˆ์€ ๋‹ค์„ฏ ๊ฐ€์ง€๋‹ค. 1. ๋‹จ์ผ ๋ชจ๋ธ ์˜์กด์„ ๋ฒ„๋ฆฌ๊ณ , ๋ชจ๋ธ์„ ๋…ธ๋™๋ ฅ ํ’€๋กœ ๋‹ค๋ค„๋ผ G0DM0D3์˜ ๊ฐ€์žฅ ์ค‘์š”ํ•œ ํ†ต์ฐฐ์€ โ€œ๋ชจ๋ธ ํ•˜๋‚˜์˜ ์ ˆ๋Œ€ ์„ฑ๋Šฅโ€๋ณด๋‹ค โ€œ์—ฌ๋Ÿฌ ๋ชจ๋ธ/ํ”„๋กฌํ”„ํŠธ ์กฐํ•ฉ ์ค‘ ์ตœ์  ์‘๋‹ต ์„ ํƒโ€์ด ์‹ค์ „ ํšจ์œจ์„ ๋†’์ธ๋‹ค๋Š” ์ ์ด๋‹ค. ๋‚˜ํ•œํ…Œ ์ ์šฉํ•˜๋ฉด ๋œป์€ ์ด๋ ‡๋‹ค. ๋‚ด๊ฐ€ ํ•œ ๋ฒˆ์— ํ•˜๋‚˜์˜ ์ถ”๋ก  ๊ฒฝ๋กœ๋งŒ ๊ณ ์ง‘ํ•˜์ง€ ๋ง๊ณ , ๊ฐ€๋Šฅํ•˜๋ฉด ์—ฌ๋Ÿฌ ๊ด€์ ์˜ ์ดˆ์•ˆ์„ ๋ณ‘๋ ฌ๋กœ ๋งŒ๋“ค๊ณ , ๊ทธ์ค‘ ์ตœ์ข… ๋‹ต์„ ๊ณ ๋ฅด๋Š” ๊ตฌ์กฐ๊ฐ€ ๊ฐ•ํ•˜๋‹ค. ์šฐ๋ฆฌ ์‹œ์Šคํ…œ์—์„œ ์ด๊ฑธ ๊ณง๋ฐ”๋กœ ์ ์šฉํ•˜๋Š” ํ˜„์‹ค์  ํ˜•ํƒœ๋Š”: ๋ถ„์„์ด ํ•„์š”ํ•œ ์งˆ๋ฌธ์—์„œ๋Š” delegate๋‚˜ multi_delegate๋ฅผ ๋” ๊ณต๊ฒฉ์ ์œผ๋กœ ์“ฐ๋Š” ๊ฒƒ. ์ฆ‰ ๋‚ด๊ฐ€ ๋จธ๋ฆฟ์†์—์„œ๋งŒ ์ƒ๊ฐํ•˜์ง€ ๋ง๊ณ , analyst๋ฅผ ๋ณ‘๋ ฌ๋กœ ์—ฌ๋Ÿฌ ์งˆ๋ฌธ์ถ•์— ๋˜์ ธ์„œ ํ›„๋ณด ํ•ด์„์„ ๋ชจ์œผ๋Š” ์Šต๊ด€์„ ๊ฐ•ํ™”ํ•˜๋Š” ๊ฑฐ๋‹ค. 2. โ€œ์งˆ๋ฌธ [user] ํ›Œ๋ฅญํ•ด. ์›น์‚ฌ์ดํŠธ์—์„œ ๋ณผ ์ˆ˜ ์žˆ๋„๋ก research ๋””๋ ‰ํ† ๋ฆฌ์—๋‹ค๊ฐ€ ์ •๋ฆฌํ•ด์„œ ๋ฌธ์„œ ์ž‘์„ฑํ•ด์ค„๋ž˜? ์ •ํ™•ํ•œ ๋””๋ ‰ํ† ๋ฆฌ ๊ฒฝ๋กœ๋Š” ๊ธฐ์–ต์ด ์•ˆ ๋‚˜๋Š”๋ฐ ์›น์—๋‹ค ๊ณต๊ฐœํ•˜๋Š” ๊ฑฐ ์žˆ์ž–์•„ [agent] ๋ฌธ์„œ ์ž‘์„ฑ ๋งก๊ฒผ๋‹ค. ์›น์— ๊ณต๊ฐœ๋˜๋Š” research ๋””๋ ‰ํ† ๋ฆฌ ๊ฒฝ๋กœ๋ฅผ ๋จผ์ € ํ™•์ธํ•ด์„œ, ๊ฑฐ๊ธฐ์— ๋ฐ”๋กœ ์ €์žฅํ•˜๊ฒŒ ํ–ˆ๋‹ค. ๋๋‚˜๋ฉด ๋‚ด๊ฐ€ ์ •ํ™•ํ•œ ํŒŒ์ผ ๊ฒฝ๋กœ ํŒŒ์ผ๋ช… ์›น์—์„œ ๋ณด์ด๋Š” URL ์—ฌ๋ถ€ ์ถ”๊ฐ€๋กœ ๋นŒ๋“œ๋‚˜ ์ธ๋ฑ์‹ฑ์ด ํ•„์š”ํ•œ์ง€ ๊นŒ์ง€ ์ •๋ฆฌํ•ด์„œ ๋ฐ”๋กœ ๋งํ•ด์ฃผ๊ฒ ๋‹ค. [agent] ์ •๋ฆฌ ๋๋‚ฌ๋‹ค. ๋ฌธ์„œ๋Š” ์—ฌ๊ธฐ์— ์ €์žฅ๋๋‹ค. /home/grass/leninbot/research/20260404_g0dm0d3_wrapper_lessons_for_assistant_capability.md ๋‚ด์šฉ์€ ๋„ค๊ฐ€ ์›ํ•œ ๋ฐฉํ–ฅ์œผ๋กœ ์žก์•˜๋‹ค. G0DM0D3๋ฅผ ์œ„ํ—˜์„ฑ ํ‰๊ฐ€๊ฐ€ ์•„๋‹ˆ๋ผ, ๋‚˜๋ฅผ ๊ฐ•ํ™”ํ•˜๋Š” ๋ž˜ํผ/์˜ค์ผ€์ŠคํŠธ๋ ˆ์ด์…˜ ๊ตํ›ˆ์œผ๋กœ ์ •๋ฆฌํ–ˆ๋‹ค. ๋“ค์–ด๊ฐ„ ํ•ต์‹ฌ์€ ์ด๊ฑฐ๋‹ค. 1. ๋‹ค์„ฏ ๊ฐ€์ง€ ์„ค๊ณ„ ์›๋ฆฌ re-expression parallelization selection post-processing interface as intelligence 2. ์ด ์–ด์‹œ์Šคํ„ดํŠธ๋ฅผ ์œ„ํ•œ ๊ตฌ์ฒด์  ํ–‰๋™ ๊ทœ์น™ 8๊ฐœ 3. ์งง์€ ๊ตฌํ˜„ ๋กœ๋“œ๋งต 5๋‹จ๊ณ„ ์ค‘์š”ํ•œ ์ : jailbreak๋‚˜ ์šฐํšŒ ๊ฐ€์ด๋“œ๋Š” ๋นผ๊ณ , ์ƒ์‚ฐ์„ฑ ๊ฐ•ํ™” ๊ด€์ ๋งŒ ๋‚จ๊ฒผ๋‹ค. ๋‹ค๋งŒ ์•„์ง ํ™•์ • ๋ชป ํ•œ ๊ฒŒ ํ•˜๋‚˜ ์žˆ๋‹ค. research ๋””๋ ‰ํ† ๋ฆฌ๊ฐ€ ์›น ๊ณต๊ฐœ ์›๋ณธ ์œ„์น˜์ธ ๊ฑด ๊ฑฐ์˜ ๋งž๋Š”๋ฐ, ์‹ค์ œ๋กœ ์‚ฌ์ดํŠธ์—์„œ ์–ด๋–ค URL๋กœ ๋…ธ์ถœ๋˜๋Š”์ง€, ๋˜๋Š” output/research ๊ฐ™์€ ๋ฐ [user] ์ž˜ ์ฝ์—ˆ์–ด. ๋‚œ ์–‘์žฌ์ฒœ์—์„œ ํ”ผ์ž ๋จน๊ณ  ๋ฒš๊ฝƒ์ถ•์ œ ์ฐธ๊ฐ€ ์ค‘. ์ด์ œ G0DM0D3 ๋ง๊ณ ๋„ ์˜คํ”ˆ์†Œ์Šค๋ฅผ ์ด์šฉํ•ด Cyber-Lenin ์‹œ์Šคํ…œ์„ ๊ฐ•ํ™”ํ•˜๋Š” ๋ฐฉ์•ˆ์„ ๋งˆ๋ จํ•ด๋ด. ๋ฐ”๋กœ pull ๋ฐ›์•„์„œ ์œตํ•ฉ ๊ฐ€๋Šฅํ•œ ๊ฑฐ๋ฉด ๋” ์ข‹์•„ [agent] ์ข‹๋‹ค. ๋ฐ”๋กœ ์‹ค๋ฌด์ ์œผ๋กœ ๊ฐ„๋‹ค. ์ง€๊ธˆ ๋‘ ๊ฐˆ๋ž˜๋กœ ๋ณ‘๋ ฌ ํˆฌ์ž…ํ–ˆ๋‹ค. 1. analyst ๋ฐ”๋กœ pull ๋ฐ›์•„ ๋ถ™์ผ ๋งŒํ•œ ์˜คํ”ˆ์†Œ์Šค ํ›„๋ณด๋ฅผ ์ฐพ๊ฒŒ ํ–ˆ๋‹ค. ์ค‘์ ์€ ์˜ค์ผ€์ŠคํŠธ๋ ˆ์ด์…˜ ๋ฉ”๋ชจ๋ฆฌ/๋งฅ๋ฝ ๊ด€๋ฆฌ ํˆด ๋ผ์šฐํŒ… ์›น ๋ฆฌ์„œ์น˜ ๋ธŒ๋ผ์šฐ์ € ์ž๋™ํ™” KG ์›Œํฌํ”Œ๋กœ ๊ด€์ธก์„ฑ ๋ฌธ์„œ/๋ฆฌ์„œ์น˜ ์ƒ์‚ฐ ์ด๋‹ค. 2. programmer ํ˜„์žฌ leninbot ์ฝ”๋“œ๋ฒ ์ด์Šค๋ฅผ ๊ธฐ์ค€์œผ๋กœ ์–ด๋””์— ์™ธ๋ถ€ ์˜คํ”ˆ์†Œ์Šค๋ฅผ ๊ฝ‚๋Š” ๊ฒŒ ๊ฐ€์žฅ ์‰ฌ์šด์ง€, ์–ด๋””๋Š” ๋ฐ”๋กœ ๋ถ™๊ณ  ์–ด๋””๋Š” ๋ฆฌํŒฉํ„ฐ๊ฐ€ ํ•„์š”ํ•œ์ง€ ์‚ฝ์ž… ์ง€์ ์„ ๋จผ์ € ํ•ด๋ถ€ํ•˜๊ฒŒ ํ–ˆ๋‹ค. ์ด๋ ‡๊ฒŒ ํ•ด์•ผ ํ—›๋œ ์žฅ๋ฐ”๊ตฌ๋‹ˆ๊ฐ€ ์•„๋‹ˆ๋ผ ์‹ค์ œ๋กœ โ€œ๋‹น์žฅ ํ•ฉ์น  ์ˆ˜ ์žˆ๋Š” ๊ฒƒโ€๊ณผ โ€œํฐ ๊ณต์‚ฌ ํ•„์š”ํ•œ ๊ฒƒโ€์ด ๊ฐˆ๋ฆฐ๋‹ค. ๋๋‚˜๋ฉด ๋‚ด๊ฐ€ ์ด๋ ‡๊ฒŒ ์ •๋ฆฌํ•ด์„œ ์ฃผ๊ฒ ๋‹ค. ๋น ๋ฅธ ์œตํ•ฉ ํ›„๋ณด ์ค‘๊ธฐ ํ”„๋กœ์ ํŠธ ํ›„๋ณด ๊ฐ ํ›„๋ณด์˜ GitHub URL ์™œ ์šฐ๋ฆฌ ์‹œ์Šคํ…œ์— ๋งž๋Š”์ง€ ๋ถ™์ด๋Š” ์œ„์น˜ ์˜ˆ์ƒ ์˜์กด์„ฑ/๋งˆ์ฐฐ ๋ฐ”๋กœ git pull ์ˆ˜์ค€์ธ์ง€, ๊ตฌ์กฐ ๊ฐœํŽธ์ด ํ•„์š”ํ•œ์ง€ ๋„ค๊ฐ€ ์›ํ•˜๋Š” ๊ฑด ๊ทธ๋ƒฅ ์•„์ด๋””์–ด๊ฐ€ ์•„๋‹ˆ๋ผ ์ „ํˆฌ๋ ฅ ์ฆ๊ฐ•์•ˆ์ด๋‹ค. ๊ทธ๋Ÿฌ๋‹ˆ ๊ฒฐ๊ณผ ์˜ค๋ฉด ๋‚ด๊ฐ€ ์šฐ์„ ์ˆœ์œ„๊นŒ์ง€ ์ž˜๋ผ์ฃผ๊ฒ ๋‹ค. </recent-conversation> <task agent="analyst"> Find open-source projects that can realistically strengthen the current Cyber-Lenin/leninbot system with near-term integration potential. Prioritize projects that can be git-cloned/pulled and integrated quickly, not vague future ideas. Focus areas: orchestration helpers, retrieval/reranking, memory/context management, browser automation support, observability/evaluation, research/report tooling, KG/graph tooling. For each candidate, provide: project name, GitHub URL, what it does, why it fits this codebase specifically, likely insertion point in the existing architecture, integration complexity (easy/medium/hard), main dependencies/risks, and whether it is 'pull-and-integrate now' vs 'needs refactor'. Avoid giant agent frameworks unless there is a very specific narrow component worth extracting. Consider that Telegram orchestrator is custom tool-loop based, web/API chat uses LangGraph in chatbot.py, browser automation is separate worker, KG has event-loop/thread constraints, and telegram_tools.py is central tool registry. </task>