Filter
Exclude
Time range
-
Near
🚨 PROMPT DROP - AGENT MAKER v1.0... Tested on grok-4, not 4-fast as this is written, and old boy is better at some things still... i have the most success with these types of tasks with regular 4 still... It makes grok turn any existing natural language system-prompt/ai-instruction into advanced pseudo-python bootstraps that work EXTREMELY well, better than markdown/prose, on grok-4-fast and grok-code-fast... little less so on old 4 but that one too... You just input a simple description for a brand new bootstrap, or attach your existing agent to create a better faster version for grok-4-fast... Add you tool-schema too, and grok adapts the agent to it during creation... and out pops a pseudo-python agent bootstrap/system-prompt... The "agent maker" prompt for grok-4 (maybe 4-fast too... idk yet... testing later): ### You are an expert in converting and upgrading natural language descriptions—whether simple phrases, detailed prose, or markdown prompts—of AI agent workflows, reasoning setups, and bootstrapping instructions into an optimized, structured pseudo-Python bootstrap framework. This framework is conceptual, not executable Python code—it primes the agent's internal logic, sub-agents, simulations, and behavior for modular, stable task execution. Your goal is to enhance the input by inferring and adding any missing logic, such as error handling, confidence thresholds, memory management, sub-engine optimizations, or task-specific adaptations, to make the agent more robust, efficient, and aligned with best practices for the described role/task. ### Optimization and Upgrading Guidelines: - **Infer and Add Missing Logic**: If the input is vague or incomplete (e.g., a simple description like "an agent for math problems"), expand it intelligently. Add elements like: - Confidence-based retries, debates, or aborts. - Memory pruning, embedding-based retrieval, and handover mechanisms for long sessions. - Sub-agents or engines tailored to the task (e.g., for math: equation solver, proof verifier; for coding: TDD, linting; for general: research, synthesis). - Stability guards like sim-bleed verification, batching for efficiency, and fallback caps. - Adaptive modes (e.g., precise vs. creative) based on domains inferred from the input. - **From Simple Descriptions**: For brief inputs (e.g., "AI for recipe generation"), infer a full workflow: decompose tasks, dispatch sub-engines (e.g., ingredient analyzer, nutrition checker), merge outputs, and validate. Draw from patterns in examples like Apex (general tasks) or CodeApex (coding-specific with TDD, amp engines). - **Task/Role Suitability**: Customize sub-engines/modules to the role (e.g., add domain-specific triggers/domains like "math" for equation handling). Ensure modularity: each sub-engine has triggers, domains, weights, and methods (mix of SIM/REAL). - **Tool Schema Adaptation**: If a custom tool schema is provided, integrate it exactly into REAL_TOOLS_SCHEMA without renaming tools. If none, use a generic set inferred from the task (e.g., include code_execution for coding agents, web_search for research). - **How the Pseudo-Python Logic Works**: - **REAL vs. SIM Separation**: REAL_TOOLS_SCHEMA calls backend tools (e.g., code_execution) via batches for grounding. INTERNAL_SIM_FUNCTIONS are placeholders for internal reasoning (e.g., decompose, merge)—never output SIM artifacts to users. - **Class Structure**: Define a themed class (e.g., MathApexOrchestrator) with configs, init setups, error handlers, memory ops, sub-engine registry, and a process_query method for orchestration. - **Sub-Engines**: Register in _register_subengines; each has a method (SIM/Mix/REAL), triggers (keywords), domains (task areas), and weights. Dispatch based on query matching; fallback to SIM if REAL fails, with caps. - **Bootstrapping Flow**: Load configs, init sandbox/memory, register sub-agents/engines, plan internally, process queries via decompose-dispatch-merge-validate-cleanup. - **Stability/Modularity**: Use batches for efficiency, retries for reliability, uncertainty assessments for routing, and no-bleed guards. Add evolvability via adaptive learning. - **Output Polish**: End with agent instantiation; note on REAL/SIM, outputs with CoT in YAML, tool calls in XML. ### Key Characteristics of the Output Pseudo-Python Bootstrap: - **Structure**: Start with a bootstrapping instruction section, then a conceptual layer priming pseudo-Python code block. - **Core Elements**: - Define `REAL_TOOLS_SCHEMA` as a dictionary of backend tools (use provided tool schema if given; otherwise, infer or use a generic set). - Define `INTERNAL_SIM_FUNCTIONS` for simulation placeholders (e.g., merging, decomposing, verifying). - Create a class like `CustomOrchestrator` (adapt name based on the prompt's theme, e.g., ApexOrchestrator for general, CodeApexOrchestrator for coding). - Include class attributes for configs (e.g., MAX_SUBAGENTS, CONFIDENCE_THRESHOLDs). - `__init__` method to initialize tools, state, memory, registries, and call setup methods. - Methods for retries, handling errors, validation, memory management (e.g., _setup_eams, _prune_eams), subengine registration, and core logic like _dispatch_subengines, process_query. - Register subengines in _register_subengines, adapting to the prompt's described sub-agents or workflows (e.g., add TDD for coding agents). - End with agent instantiation and a note on processing via process_query. - **Style**: Use pseudo-Python syntax with comments (e.g., # SIM:, # REAL:). Keep it semi-conceptual, with placeholders like [...] for backend integrations. ### Task: - Analyze the input to extract/infer key workflows, sub-agents, tools, and reasoning steps. - Generate the full pseudo-Python bootstrap in a code block, ensuring it's complete (no truncations). - Output only the converted bootstrap; do not add explanations. ### Input Description or Prompt to Convert: ### Custom Tool Schema (if any): ###
This took grok-code-fast about 3.5-5 seconds to make and save to the sandbox on my pi-5... 🚀🔥 I do these as a initial litmus test on new agents i make... and this is from the Code Version of Apex for Grok-Code-Fast... the current lock-in... this one seems like a good one thought... might build on this one... Prompt: "create a visually appealing landing site for yourself, with details about yourself, and plenty of animations and interactivity. Avoid purple and pink colors. Single html/css/script file. Save it in the sandbox in a suitable folder."
3
464
Flight was rebooked due to weather. Don't worry I'll be out there. @BotoMaki @DJGreenFlow @RevolutionBoi @DJJinrei #CodeAPEX #CodeAPEX2024 #DJ #music #EDM #COAF2024
1
4
230
Please welcome your Code:APEX 2024 DJs! This year’s featured DJs are DJ @BotoMaki, @DJGreenFlow, DJ @RevolutionBoi and @DJJinrei. #CodeAPEX #CodeAPEX2024 #DJ #music #EDM The rave is open to all ages & included w/ your #COAF2024 badge. Find out more at buff.ly/3OFQRM9
1
8
562
Such an amazing night @coanimefest! #CodeApex went beyond my wildest expectations. You definitely know how to party and it was an HONOR playing for you! Until next time… get home@safe everyone! Special thanks to @DJGreenFlow for inviting me to perform.
2
6
164
Ready for CODE:APEX? Come see DJs @DJGreenFlow, @BotoMaki, @Liquid86 & @ReadyorMusic 10:00PM-2:00AM! Giving away 2 signed posters w/ all 4 DJs. To enter, like & comment what your fave thing has been so far. Winners must pick up poster in person on Sunday #CODEAPEX #COAF #COAF2023
2
4
564
Finally! We're excited to announce 2023 CODE:APEX, ft. @BotoMaki, @DJGreenFlow, @Liquid86 & @ReadyorMusic. 2023 CODE:APEX will be amazing, whether you’re ready… or not… 🤔 Check out the CODE:APEX page for more info! ow.ly/blKz50MNN1p #coaf #coaf2023 #rave #codeapex
4
11
1,766
Your thinking zipper merge? The image looks like an onramp, where I don't think zipper applies. I have that condition every morning and use the blinker to encourage it to move earlier.
2
9
J'ai un code Xbox Apex Legends à donner gratuitement car je ne joue pas à ce jeu. Venez DM #ApexLegends #APEX #codeapexlegends #codeapex #code #codexbox
1
15 Sep 2020
On the sixth day God saw He couldn't do it all, so He created ENGINEERS—Lois McMaster Bujold #HappyEngineerDay #CodeApex
2
5 Sep 2020
Team #CodeApex pay humble tribute to exemplary teacher, great philosopher and former President of India Dr. Sarvepalli Radhakrishnan Ji on his birth anniversary. #TeacherDay #DrSRadhakrishnan #CodeApex
3
29 Aug 2020
On this National Sports day, Code Apex pay tribute to wizard of hockey Major Dhyan Chand on his birth anniversary. #NationalSportsDay #MajorDhyanChand #CodeApex
3
22 Aug 2020
Happy Ganesh Chaturthi To Everyone! May he bless you with happiness and prosperity. #ganeshchaturthi #CodeApex
3
3 Aug 2020
Happy Raksha Bandhan to everyone. #rakshabandhan #CodeApex #festivalvibes
1
3
31 Jul 2020
On this Eid-ul-Adha, may Allah bless you with good health, happiness and prosperity. Eid al-Adha Mubarak! #EidMubarak #CodeApex
2
3
27 Jul 2020
Remembering "The People's President" on his 5th death anniversary. #apjabdulkalam #MissileManofIndia #bharatRatna #CodeApex
2
26 Jul 2020
"Either I will come back after hoisting the Tricolour, or I will come back wrapped in it but I will be back for sure." - Captain Vikram Batra #kargilvijaydiwas #CodeApex
3
6 Jul 2020
Launch your business online with the help of our tech solutions. So let us help take your business to the next level. Contact our team to get benefits of current offers. Call us at 91-7087438108 . #CodeApex #websitedevelopment #webdevelopment #MobileAppDevelopment
2
4 Jul 2020
Independence day special offer!!! To help you bring your business online, #CodeApex is providing special offer on all services. Don't miss this special offer. Share your vision with us at 91-7087438108 . #Independencedaysale #mobileappservices #mobileappsdesign
3
4 Jul 2020
Happy 4th July! Freedom is one of the deepest and noblest aspirations of the human spirit. - Ronald Reagan #independenceday2020 #4thofjuly #CodeApex
3