Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-03-03 · source-backed
Each link below shares sources, entities, or timing with this story.
Research paper validates a three-tier architecture for AI coding context: hot-memory constitution (conventions + retrieval hooks), specialized domain-expert agents, and cold-memory on-demand specs. Projects with AGENTS.md showed 29% reduction in median runtime and 17% in outpu...
Hot memory + 19 specialized agents + cold knowledge base, evaluated across 283 sessions on a 108K-line codebase. Open-source companion repo. Blueprint for scaling agentic coding with structured context engineering. (arXiv 2602.20478)
First defense against MCP Tool Poisoning Attacks using a Decision Dependence Graph that correlates LLM attention with tool invocation decisions. 97%+ detection accuracy with zero token overhead. Key insight: behavior-level defenses are fundamentally ineffective against TPA bec...
157 tasks from 27 repos. Best resolved rate: 29.94%. Agents cause scope creep and regressions by diverging from user intent. Practical implication: agents need explicit scope constraints. arXiv 2509.22237 ---
39. CoSAI MCP Security — CoSAI 40. Agent Protocol Security — arXiv 41. Elastic MCP Attack Vectors — Elastic Security Labs 42. NVIDIA Sandboxing — NVIDIA 43. MCP Best Practices — Phil Schmid 44. Sonnet 4.6 Benchmarks — VentureBeat ---
In a GitHub Copilot SDK setup, an asynchronous memory-curator agent got read-only tools to check candidate memories against the current state before saving them. Pass rate on CLBench rose to 73% from 39% (arXiv 2609.11060). Queries per question fell to 4.7 from 8.8, and task-a...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.