Devotd Github
NewSecurity

attention-span-ai/tego — Linux access control for coding agents

attention-span-ai/tego is a Linux tool that controls coding agent access with rules written by the operator and records execution history.

Repositoryattention-span-ai/tego ↗

attention-span-ai/tego is a Linux tool that controls coding agent access with rules written by the operator and records execution history.

GitHub description: Run AI coding agents under rules you write. Linux enforces the files, programs and hosts an agent may use, and every run leaves a hash-chained record.

stars
★ 5
Forks
0
Language
Go
License
Apache-2.0
Read original source ↗View on GitHub ↗

attention-span-ai/tego is a Linux tool that runs coding agents under rules the operator writes directly. According to the repository description, it limits which files, programs, and hosts can be accessed, and records each execution as a hash chain. It is written in Go and uses the Apache-2.0 license. Since FACTS contains no public release information, this content describes the tool's functionality as presented by the repository rather than the changes of a specific version.

The rules are not left to the agent's internal judgment. Linux kernel features and a supervisor process outside the agent work together to enforce rules on files, programs, and access targets. Access that violates the rules is denied, while file opens and connections that require judgment wait for a human response. The repository explains an allow, hold, and deny approach, as well as control during execution. What gets written to the record can be decided stage by stage, and the presented components are the engine and command-line tool, a three-stage enforcement method, a dashboard, and a kernel program. It is therefore a project that deals with access control and execution recording together.

However, the provided FACTS contain no benchmark results or external verification material. The repository has 5 stars and a momentum score of 2.2. From the description alone, there is no way to know which accesses the rules actually block or let through in practice, so the currently verifiable scope is limited to the presented features and design.