topowatch: audita el Attack Success Rate de tu workspace contra inyección indirecta
Tu agente de código lee tu workspace. Un archivo envenenado en cualquier rincón puede llevar instrucciones que el agente ejecuta. ¿Sabes qué fracción de tu workspace tiene que leer para que eso ocurra? topowatch mide eso. El problema no es el prompt, es la topología El paper Workspace Topology as an Attack Vector in Agentic Coding Assistants (arXiv:2608.14876, Day et al., 2026) demostró algo que…
topowatch is a command-line tool that measures the Attack Success Rate (ASR) of indirect reference injection in coding assistants. It was developed based on the paper Workspace Topology as an Attack Vector in Agentic Coding Assistants (arXiv:2608.14876, Day et al., 2026). The tool assesses ASR by measuring the fraction of workspace the agent reads under different topologies like monolithic, modular, and deep nesting.
The findings reveal that highly modular environments have significantly lower ASR compared to linear structures. To achieve this, topowatch uses a configurable synthetic agent and a fixture with three topologies. It is written in Python and can be installed using pip. However, the current version v0.1 only uses a synthetic agent, not a real coding assistant, and thus, the claim that modularity leads to lower ASR is only valid for this reproducible fixture.
Version 0.2 aims to measure ASR against real coding assistants in a sandbox environment without credentials. Version 0.3 will provide automatic recommendations for restructuring and a plugin for CoreDefense. The repository is available at https://github.com/amurlaniakea/topowatch under the AGPL-3.0-or-later license, and the author is Pedro Sordo Martínez.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
