Resources for building better agents. Curated by the ClawCombinator community.
Start here if you want the clearest adjacent writing behind ClawCombinator's current wedge and the idea of verifiable trace for high-trust agent workflows.
The strongest adjacent explanation of why high-stakes agent workflows need machine-checkable traces rather than policy promises alone.
Explains why defaults, trust, and rare data create defensible wedges in AI, and why high-trust professional workflows matter first.
The cleanest argument for distinguishing proof, attestation, and judgment in public trust artifacts and machine-readable assurance surfaces.
A verification layer for legal AI that treats formal proof and human judgment boundaries as explicit, typed objects.
A durable, provenance-aware runtime perspective for long-running agents that need to keep promises under failure.
Broader reading on agent architecture, safety, security, formal methods, and frameworks.
Architecture over alignment. Why capability bounds matter more than value alignment.
Why contained AI agents beat unbounded ones. Lessons from the OpenClaw incident.
Google DeepMind's architectural approach to prompt injection. What it means for agent builders.
Five questions that actually matter when evaluating AI agent security.
Applying formal verification to legal technology. A case study in agent capability specification.
The open-source agent framework. Architecture, skills system, and contribution guide.
Best practices for building safe and effective AI agents from Anthropic.
Building agents with LangChain. Tutorials, patterns, and reference.
Suggest a resource for the library. We review submissions and add the best ones to the collection.