Checks Chinese copy against a checklist that removes the telltale tone of machine-written text.
bojieli/ai-agent-book
A companion skill collection for the book Understanding AI Agents, covering agent design, evaluation, context, memory, tools and training.
- 30 skills
- 58.6k stars
The 5 most installed
Ordered by installs on skills.sh- ai-style16 installs
- pptx14 installs
Generates a PowerPoint deck from a paper, outline or structured text.
- data_analysis4 installs
Runs auditable math and descriptive statistics on data you have already confirmed.
- coding3 installs
Writes and runs Python code to check that scripts and their output are correct.
- research2 installs
Finds traceable facts, data and sources using real search tools.
The other 25 skills
- triage2 installs
Breaks a complex task into stages and picks the right skill for the next step.
- writing2 installs
Turns verified facts and results into a finished piece matching your audience and format.
Builds an evaluation system for your agent, covering metrics, test environments, rubrics, failure analysis and regression checks.
Helps your agent learn from its own run history and improve itself through experience, prompts, code or model updates.
Adds a status bar to your agent so it tracks time, tool calls, TODOs and errors instead of looping or drifting.
Builds agents that react to emails, webhooks, timers and alerts, and handle interruptions while tools are running.
Turns failed production runs into training data, from failure analysis and preference pairs to LoRA training and validation.
Guides building a coding agent that reads and writes code, runs tests and fixes itself, plus the guardrails around it.
Builds agents that see the screen and use mouse and keyboard to operate desktop or browser software.
Shrinks an agent's context when tool results overflow the window, using layered compression strategies and sub-agents.
Helps write agent system prompts, structure context and author on-demand skills, including defenses against prompt injection.
Suggests a skill about training a model on curly-quote text, based on its name only.
Hardens an agent against failures like loops, dropped streams, rate limits and context overflow, with retries and circuit breakers.
Designs agent evaluation tasks and datasets, including validators, difficulty tiers, leak prevention and long-term upkeep.
Organizes and retrieves knowledge beyond flat text chunks, using structured indexes, file systems and agentic retrieval.
Speeds up agent inference and cuts cost by keeping the KV cache intact and avoiding context patterns that break it.
Fixes agent loops that stop too early, fake success or never end, using validators and a proposer-reviewer setup.
Helps connect third-party capabilities to an agent, choosing between MCP and a skill hub and reviewing token cost and security.
Designs and evaluates an agent's user memory, covering storage formats, frameworks, compression and privacy.
Helps decide between one agent and many, then design the topology, communication and failure handling for multi-agent systems.
Guides choosing and running post-training for agent models, from mid-training and SFT to RL, DPO and distillation.
Builds and tunes a RAG retrieval pipeline, covering chunking, embeddings, hybrid search, reranking and retrieval metrics.
Helps you design reward functions for RL training and avoid reward hacking.
Helps you design and describe tools for agents, including parameters and safety checks.
Lets the agent find and load tools on demand instead of loading them all upfront.