Posts tagged "agents"
7 posts
Project Glasswing Expanded: What It Means for Your Agent Prompts
Anthropic expanded Project Glasswing to ~150 orgs across 15+ countries. Here's what their stated threat model tells you about hardening your Claude agent system prompts.
Build the Eval Stack That Survives Production with Claude
Claude writes the code—you write the assertions. Build a three-layer eval stack: Claude-generated property tests, human invariant checks, and a CLAUDE.md CI gate.
Claude Code Dynamic Workflows: How to Actually Drive Them
Dynamic Workflows let Claude Code fan work out across parallel subagents. Here's what they actually are, how to turn them on, and how to drive a branching one.
Claude Managed Agents as MCP Tools: Build a Persistent Agent Pipeline
Anthropic's new cookbook wraps the CMA Sessions API as a full MCP server. Here's the architecture, the SSE shim, and how to wire it into Claude Desktop.
Real-Time Tools Break Claude Agents — Build the Audit Guard
A viral clock failure shows a real agent safety bug: stateful tools destabilize Claude's reasoning loop. Reproduce it, then build the assertion guard that stops it.
Prompt Injection Hit My Claude App — Here's the Guard Layer I Built
A real-world prompt injection attack on a Claude app, reproduced in a sandbox, with three concrete defenses: input sanitization, integrity checks, and system prompt hardening.
Give Claude Code a Self-Maintaining Wiki That Survives Session Resets
Wire a Markdown wiki into Claude Code so the agent reads context on startup and writes what it learned on exit — no vector DBs, no cloud services.
← Back to blog