MANTA rewires multi-agent communication topology mid-task, scoring 74.0 across five benchmarks.

Shared entity: Most / Same source domain / Shared topic / Earlier coverage

Microsoft SkillOpt Trains Markdown Skills for Frozen LLMs. 52/52 Benchmark Wins. Your CLAUDE.md Can Be Optimized Like a Neural Network.

Both cover Most; reported by the same outlet (arxiv.org); overlapping topics (agent, apply, benchmark).

Your CLAUDE.md holds 36.2% of the time, and that's the best case

Both cover Most; reported by the same outlet (arxiv.org); overlapping topics (agent, benchmark).

Cursor Rebuilt SQLite With Agents for 1,339 Dollars

Both cover Most; reported by the same outlet (arxiv.org); overlapping topics (agent, execution).

EvoAgentBench isolates whether agents actually turn past sessions into reusable procedures.

Both cover Most; reported by the same outlet (arxiv.org); overlapping topics (agent, benchmark).

SWE-CI: New benchmark reveals agents fail at code maintenance.

Both cover Most; reported by the same outlet (arxiv.org); overlapping topics (agent, benchmark).

Same source domain / Shared topic / Tension

Typed memory retrieval doubled a long-horizon agent's win rate.

Reported by the same outlet (arxiv.org); overlapping topics (agent, baseline, bounded, design); pushes against this story (versus).

Shared entity: Most / Same source domain / Earlier coverage / Downstream implication

43 Attacks Can Fully Compromise a Self-Hosted AI Agent

Both cover Most; reported by the same outlet (arxiv.org); earlier Most coverage from 2026-07-22.

Shared entity: Most / Shared topic / Earlier coverage / Tension

AWS published a production MCP server build with two-layer JWT auth.

Both cover Most; overlapping topics (agent, deployment); earlier Most coverage from 2026-07-10.