Bojie Li

Design a running status summary that keeps an agent aware of tasks, tool call counts, time, errors, and its environment. The skill covers updating that summary and balancing cache reuse against the buildup of outdated information.

• Addressing repeated tool calls and retry loops
• Keeping TODO progress and original goals visible
• Choosing how to update runtime status messages

Write small, reproducible Python scripts and run them in a sandbox using execute_python. Check the script's logic and output, then report the actual results or errors while clearly distinguishing executed results from code that has not been run.

• Checking a Python script's logic
• Confirming actual output from a short script
• Reproducing an error with minimal code

Design a process for AI agents to learn from execution records. Evaluate outcomes, actions, and quality, choose whether to update knowledge, instructions, code, or model parameters, and define independent validation, release checks, and rollback procedures.

• Turning repeated agent failures into reusable guidance
• Designing evidence-based reviews of execution records
• Converting browser actions into validated workflows

Design knowledge systems that preserve document structure and connections across sources. Choose retrieval methods, organize conversation memory, and plan evidence-based updates, review workflows, and access controls for information that scattered text searches cannot adequately cover.

• Designing searches that combine information across documents
• Organizing and retrieving long conversation histories
• Establishing knowledge update and review workflows

Design and improve retrieval for RAG, which supplies external knowledge to AI answers. Cover document splitting, semantic and keyword search, combined result ranking, and quality measurements to identify where relevant information is being missed.

• Building search for a knowledge base
• Comparing semantic, keyword, and hybrid search
• Diagnosing missed search results

Estimate GPU memory, training step time, checkpoint costs, and recovery overhead for training or reinforcement learning. Compare state partitioning, recomputation, offloading, and device counts to assess whether a job can finish within its deadline.

• Check whether training can finish by a deadline
• Compare memory-saving methods and their time costs
• Plan checkpoint intervals and recovery overhead

Compare batching, request scheduling, and memory management for a single model serving instance. Assess concurrent request capacity, response times, generation speed, and GPU time per result that meets the required quality and deadline.

• Comparing batching and request scheduling options
• Checking memory capacity for concurrent requests
• Assessing cache reuse and offloading tradeoffs

Estimate how model architecture, precision, and context length affect weight storage, generation state, computation, and memory traffic. Compare models and attention designs with separate resource estimates for input processing and token generation.

• Compare memory needs across model architectures
• Estimate resource costs of longer contexts
• Compare attention designs and expert-based models

Plan how to split AI models and workloads across multiple GPUs. Compare memory needs, computation, communication, and waiting time to estimate completion times and choose a layout suited to response time, throughput, or cost goals.

• Running models that exceed a single GPU's memory
• Comparing ways to split work across GPUs
• Estimating communication costs within and between servers

Plan when an agent’s tool environments are created, kept available, released, and restored. Estimate peak concurrency, resource use during work and waiting, total task duration, and cost across model calls, tool execution, validation, and retries.

• Estimating peak concurrent agent environments
• Comparing environment retention and recreation
• Assessing cold starts and shared environment pools

Model ongoing requests and multi-step agent work to estimate capacity, latency, and cost per completed task. Break execution into stages, account for queues and retries, and produce workload tables with average and peak resource budgets.

• Estimate capacity for ongoing AI requests
• Analyze queues and tool waits in agent workflows
• Compare resource costs per completed task

Estimate whether a model execution fits the available hardware and identify likely performance limits. Separate memory capacity, computation, data movement, and sequential waiting to produce a table of assumptions, ideal minimum times, unknowns, and measurement priorities.

• Checking whether a model execution fits the hardware
• Estimating ideal computation and data transfer times
• Assessing the overall benefit of a local optimization

Compare computing devices for a defined AI model workload using memory capacity, computing speed, data transfer, energy use, and cost. Build a comparison that separates predicted performance from measurements and identifies the constraints that determine the choice.

• Choosing hardware for a specific AI model
• Checking whether a model fits in device memory
• Explaining gaps between advertised and measured performance

Compare ways to distribute AI inference across multiple instances, including replicas and separate input-processing and generation stages. Estimate capacity, throughput, delays, state-transfer costs, and recovery overhead to support routing and scaling decisions.

• Compare replicas with separate inference stages
• Plan routing around compute capacity and cached state
• Evaluate expert placement and load imbalance

Compare running models or tasks on a device, nearby edge infrastructure, or in the cloud. Break down interaction delays, data transfers, energy use, costs, and failure recovery to assess deployment choices for multimodal and real-time agents.

• Comparing device, edge, and cloud execution
• Estimating interaction delays on unstable networks
• Assessing battery use and sustainable processing speed

Estimate cross-server traffic, network bottlenecks, and synchronization waits for AI training or inference. Map communication onto physical network paths and assess how topology or communication changes affect the time needed to complete the job.

• Find communication bottlenecks in distributed AI workloads
• Compare network topologies and communication paths
• Estimate synchronization waits under concurrent traffic

Analyze GPU operation timings and execution traces to find costly data access, repeated computation, launch overhead, and synchronization waits. Check whether improving an individual operation actually reduces the time needed to complete a full request.

• Investigating slow GPU operations
• Assessing whether combining operations reduces data access
• Checking pipeline dependencies and buffer reuse

Build or revise private, cited Markdown notes from the user's WeChat history in the current agent session. Use for a maintained personal wiki or incremental notes; use greenbubbles-context for a one-off lookup or summary.

The English summary is being prepared. Showing the original description.

Install or diagnose the local GreenBubbles CLI and read-only WeChat access. Use for a missing CLI, failed database access, source selection, or initial owner-operated key capture.

The English summary is being prepared. Showing the original description.

Find chats by name, read or search WeChat messages with the local GreenBubbles CLI. Use for bounded live or snapshot queries; route setup problems to greenbubbles-setup and maintained notes to greenbubbles-personal-memory.

The English summary is being prepared. Showing the original description.