Agents and harnesses for advanced learners
288 resources · page 8 of 9
Agent Skills specification
Creator not supplied; see original source.
Agent Skills specification: format for SKILL.md skill directories, metadata and progressive disclosure of instructions.
aider docs: Edit formats
Creator not supplied; see original source.
Aider documentation on edit formats (whole, diff, udiff) and why different models need different formats.
aider docs: Repository map
Creator not supplied; see original source.
Aider documentation on its repository map, a concise tree-sitter summary of a codebase given to the LLM as context.
Anomalib
Creator not supplied; see original source.
Anomalib documentation: deep-learning library of anomaly detection algorithms with training, benchmarking and deployment tools.
How we built our multi-agent research system
by Anthropic
Real production lessons on orchestrator/sub-agent design and its costs.
Writing effective tools for AI agents
by Anthropic
How to design, evaluate and improve the tools an agent calls.
Browser Use
Inspect a Python library for model-driven browser interaction.
Building a C compiler with a team of parallel Claudes
Creator not supplied; see original source.
Anthropic engineering case study of building a C compiler with many Claude agents working in parallel.
CEPEJ assessment tool for AI in judicial systems
Creator not supplied; see original source.
Council of Europe CEPEJ assessment tool (PDF) operationalising its ethical charter on AI in judicial systems.
A harness for every task: dynamic workflows in Claude Code
Creator not supplied; see original source.
Claude blog post on dynamic workflows in Claude Code, where it writes and orchestrates its own multi-agent harness.
Code Mode: the better way to use MCP (Cloudflare Blog)
Creator not supplied; see original source.
Cloudflare post on Code Mode, where agents write TypeScript against MCP APIs instead of calling tools directly.
Don't Build Multi-Agents (Cognition)
Creator not supplied; see original source.
Cognition post by Walden Yan arguing for shared context and single-threaded agents over naive multi-agent designs.
Don't Build Multi-Agents
by Cognition
The counter-argument: context sharing beats splitting work across agents.
Multi-Agents: What's Actually Working (Cognition)
Creator not supplied; see original source.
Cognition follow-up post on which multi-agent setups work in production and where single agents still fit.
Effective context engineering for AI agents (Anthropic)
Creator not supplied; see original source.
Anthropic post on context engineering for agents: treating context as finite, with compaction, memory and tool selection.
Engineering fast, performant AI agents through fine-tuning (Decagon)
Creator not supplied; see original source.
Decagon engineering post on using several fine-tuned smaller models to cut latency in customer-service agents.
DIPY
Creator not supplied; see original source.
DIPY documentation: Python library for diffusion MRI and 3D/4D neuroimaging processing.
Direct Preference Optimization
Creator not supplied; see original source.
arXiv paper (2023) introducing Direct Preference Optimization (DPO), preference training without a separate reward model.
DSPy docs: Optimization overview
Creator not supplied; see original source.
DSPy documentation on optimisers such as MIPRO and GEPA for tuning prompts and weights programmatically.
Effective harnesses for long-running agents
Creator not supplied; see original source.
Anthropic post on harnesses for agents that work across many sessions, using an initializer agent and incremental progress.
FINOS AI Governance Framework
Creator not supplied; see original source.
FINOS AI Governance Framework: open-source controls for adopting AI in financial services, mapped to risks.
How we built our multi-agent research system (Anthropic)
Creator not supplied; see original source.
Anthropic post on how it built Claude's multi-agent Research system with an orchestrator and subagents.
Flower documentation
Creator not supplied; see original source.
Flower documentation: open-source framework for federated learning, simulation and deployment.
Google Agent Development Kit
Creator not supplied; see original source.
Google Agent Development Kit (ADK) documentation for building, evaluating and deploying multi-agent systems.
Grok 3 / DeepSearch
Creator not supplied; see original source.
xAI announcement of Grok 3 Beta, describing its reasoning modes, DeepSearch and benchmark results.
Harness design for long-running application development
Creator not supplied; see original source.
Anthropic post on harness design for agents that build applications over long runs.
How we contain Claude across products
Creator not supplied; see original source.
Anthropic engineering post describing the containment architecture used to isolate Claude across its products.
12-Factor Agents
Principles for agents that hold up in production.
Intercom and Fin developer platform
Creator not supplied; see original source.
Intercom developer platform: APIs, SDKs and integration guides for Intercom and its Fin AI agent.
JAX documentation
Creator not supplied; see original source.
JAX documentation: Python library for accelerator-oriented array computing with automatic differentiation, JIT compilation and sharding.
Kimi K2.5: Visual Agentic Intelligence
Creator not supplied; see original source.
arXiv paper (2026) Kimi K2.5 technical report on visual agentic intelligence.
Kimi K2: Open Agentic Intelligence
Creator not supplied; see original source.
arXiv paper (2025) Kimi K2 technical report: MuonClip optimizer and agentic post-training.