Every processed story in chronological order, with the newest coverage first. Filter by tag, source, or score to drill in.
SecureClaw is the first architecture evaluated across AgentDojo, AgentLeak, and ASB in a common harness that closes both the plaintext-exposure and unauthorized-action boundaries simultaneously, rather than trading one surface for the other.
The framework replaces the standard text-concatenation bottleneck in multi-agent synthesis with direct KV cache consumption, cutting time-to-first-token by up to 11x while preserving or improving task accuracy across diverse benchmarks.
GTBP directly addresses the two key failure modes of existing context adaptation methods — inaccurate credit assignment and lack of convergence guarantees — in multi-LLM agentic pipelines, providing both theoretical stability proofs and empirical gains across three benchmarks.
Bastion removes the environment-conflict bottleneck that prevents running multiple coding agents simultaneously by giving each agent its own fully isolated VM, enabling true parallel agent workflows on self-hosted infrastructure.
QodFlow treats AI agents as first-class participants in a shared work board, giving them a structured mechanism to pause on irreversible decisions and hand off to humans — rather than requiring a separate integration layer or chatbot interface.
The checklist and `mcp-probe` score expose a class of MCP server defects — ambiguous tool descriptions, missing argument metadata, and silent `initialize` drops — that pass standard connectivity tests but cause agents to pick wrong tools or hallucinate arguments at runtime.
MCP360 Universal Gateway consolidates what would otherwise require dozens of separate API integrations into a single MCP connection, letting AI agents discover and execute a broad set of external tools without per-service setup.
The reasoning override support for subagents closes a configuration gap that previously prevented per-subagent model and variant customization, while the recursive-deletion guard removes a data-loss risk tied to skill removal.
Batta shifts security review to the plan phase of AI agent workflows, addressing design flaws before code is generated rather than catching them at PR time or post-deployment.
The project offers a concrete, tool-checkable alternative to same-model self-verification, grounding agent reliability in deterministic external signals rather than the model's own re-reads.