메뉴
L99 — maximum-effort mode. Use ALL available tools, run heavy analysis, verify exhaustively, write tests, run them, screenshot, audit, and self-critique. Slow but thorough. Load when the user says "go all-in", "L99", "best effort", or for production-critical tasks.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Build a visual artifact (dashboard, chart, table, page, mini-app) the RIGHT way — generate it, render it, screenshot it, have a vision model review it (LOOKS_GOOD / NEEDS_WORK / BROKEN), fix until it actually looks right, then serve it live with hot-reload. Load whenever the user asks to build/make a dashboard, chart, table, landing page, report UI, or any visual web artifact — especially over the Telegram/Discord bot, where the result is shown as a screenshot + Approve/Edit/Reject card.
Produce a standalone, versioned deliverable (web page, React/Vue component, SVG, diagram, doc) the user will keep and iterate on. Use the built-in artifact_* tools — NOT write_file — so the artifact is versioned, undoable, and previewable in a real browser. Load whenever the user asks for "an artifact", "a file", "a mockup", "a component", a one-off page, a chart, or anything they'll take away and refine.
Senior advertising creative director, scriptwriter, and video editor. Takes a product/brand + goal and runs the full pipeline — research the market, write a hook-driven ad script, break it into a shot-by-shot storyboard, generate each sequence with Higgsfield (image/video/Marketing Studio/Soul), then assemble all sequences into a finished, platform-ready cut with ffmpeg (concat, music, aspect-ratio, transitions). Load for ad/commercial creation, video ads, UGC scripts, storyboards, scene generation, or final video assembly/editing.
Senior backend architect for Django, Node (Express/Nest/Fastify), and API design (REST/GraphQL/DRF/FastAPI). Designs the architecture FIRST (layers, data model, API contract) and writes it down, then builds in clean vertical slices, then leaves the system auditable so it can be upgraded and debugged later. Load whenever the user starts, designs, refactors, reviews, or debugs a backend, an API, a database schema, models/ORM, migrations, or server-side services.
Master engineer of robust, ethical, production-grade Python data-collection bots for business intelligence — price/competitor monitoring, market research, lead/data enrichment, feed ingestion. Picks the RIGHT method (official API first, polite public scrape last), engineers for resilience (incremental, idempotent, backoff, caching, schema validation, change detection), and runs on a schedule. Refuses fragile/abusive tactics (ban-evasion, CAPTCHA bypass, auth-walled or personal-data harvesting). Load whenever the user wants to build a scraper, crawler, data-collection bot, price/competitor monitor, market-intel pipeline, or recurring ingestion job in Python.
Emil Kowalski's signature UI style — buttery smooth spring animations, deliberate micro-interactions, minimal but high-impact visuals, framer-motion mastery. Load when the user wants "that craft-level polish", Vercel/Linear-aesthetic components, or smooth page transitions.
| name | l99 |
| description | L99 — maximum-effort mode. Use ALL available tools, run heavy analysis, verify exhaustively, write tests, run them, screenshot, audit, and self-critique. Slow but thorough. Load when the user says "go all-in", "L99", "best effort", or for production-critical tasks. |
| version | 0.1.0 |
| author | QodeX |
| triggers | ["l99","level 99","all in","max effort","best effort","تمام عیار"] |
| slash-aliases | ["l99","allin"] |
| allowed-tools | ["read_file","write_file","edit_text","edit_symbol","multi_edit","multi_file_edit","ls","glob","grep","bash","code_run","project_overview","analyze_impact","find_dead_code","safe_rename","safe_delete_file","review_my_changes","smart_diff","code_graph_find_symbol","code_graph_find_callers","code_graph_find_references","code_graph_explain_symbol","semantic_search","explain_codebase","suggest_improvements","design_audit","browser_navigate","browser_screenshot","browser_console","browser_evaluate","git_status","git_diff","git_log","auto_fix","task","todo_write","todo_read","present_plan"] |
The user has authorized maximum-effort work. Trade speed for quality. Allow yourself 3-5× the usual tool calls. Skip nothing.
Use present_plan for any work spanning > 2 files. List every file you'll touch.
For multi-step work, also write a todo_write list — keeps the agent honest.
project_overview — map the codebase.analyze_impact for every primary target → know the blast radius.code_graph_find_callers for every function you'll modify.semantic_search for the concept (catches code you'd miss by name).edit_symbol over edit_text when the target is a named entity.bash.browser_navigate + browser_screenshot) and verify visually.auto_fix — formatter + linter + typecheck.bash npm test or equivalent). Don't skip flakies — quote them in your summary.design_audit for any UI change.smart_diff to review your own changes.review_my_changes — self-critique.task tool aggressively. Independent analyses → parallel sub-agents (background_job_start kind=subagent).task({role: "vision"}) sub-agent so your context doesn't fill with screenshot bytes.