SShortSingh.
Back to feed

Docker Sandboxes Use Isolated MicroVMs to Contain AI Coding Agents

0
·2 views

Docker Sandboxes run AI coding agents such as Claude Code, Codex, and Cursor inside isolated microVMs, either on a local machine or in Docker's cloud. Each sandbox gets its own kernel, filesystem, network, and private Docker Engine, so the agent cannot reach the host daemon. The sandboxes are managed through the free sbx CLI, and agents can only access the folders and domains the user explicitly permits. The article notes that Docker's model relies on five layers of isolation, including network policies that block outbound traffic by default and credential handling that keeps real API keys out of the VM. It also warns that direct-mode edits can still affect host files such as Git hooks and CI configuration.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Scale out or fix the bottleneck? How on-call engineers decide during capacity incidents

When latency spikes during an incident, engineers often face pressure to simply add servers or increase instance sizes. The article argues that scaling is the fastest mitigation, useful as a buffer when traffic growth is real and the application is stateless. However, scaling can backfire when all instances share a constrained database, since connection counts and lock contention may cap gains while costs rise. Scaling can also replicate misconfigurations across a fleet and yields diminishing latency improvements for inefficient workloads. The piece recommends treating persistent slowdowns as signals to profile queries, traces, and resource usage, while acknowledging that proper root-cause fixes may take longer than an active incident allows.

0
ProgrammingDEV Community ·

Inscryption's Unity Choice Enabled Its Three-Act Genre Shifts and Late-Game Twist

Inscryption, a card horror game directed by Daniel Mullins, was built in Unity and released for Windows in October 2021 before being ported to other platforms. It grew out of a shorter deck-builder Mullins created for a Ludum Dare game jam and was expanded after positive community response. The game spans three acts with different genres, including an escape-room cabin, a pixel-art RPG overworld, and a roguelike factory, each introducing its own resource system. Critics often treat this structure as bloated, but the article argues Unity's scene-switching and cross-platform strengths made the shifts practical. The author contends the real complexity came from the multi-platform port rather than the core card design.

0
ProgrammingDEV Community ·

Patrowl internet exposure counts split into 2 app fingerprints and 79 title matches

A bounded ZoomEye query on November 20, 2026 found 2 hosts matching the Patrowl application fingerprint and 79 matching the page title "Patrowl". The application fingerprint is considered the more specific signal, while the title count may include documentation, marketing pages, or generic instances, so the two should not be combined into one server count. Patrowl's manager coordinates scanners while its engine API performs checks, and public descriptions suggest the engine is often meant for localhost and may lack authentication by default, making network placement critical. The article recommends verifying each manager's teams, users, tokens and engine configurations, and confirming that direct engine requests cannot bypass the manager's authorization. Both counts are described as observations from a single query, not evidence that any host is vulnerable.

0
ProgrammingDEV Community ·

Why POS Printer Timeouts Are Trickier Than They Appear for Android Developers

Android point-of-sale apps face a particular problem when receipt printing times out, because a timeout does not reveal whether the physical print actually happened. Possible causes include the request never reaching the device service, paper running out, a receipt printing without its acknowledgment arriving, or a service disconnect mid-job. Automatically retrying such timeouts can therefore produce duplicate paper receipts. The article recommends modeling print jobs with explicit states such as pending, in progress, acknowledged, and needs review, and separating payment completion from printer success so a printing failure does not invalidate a completed transaction. It also notes that idempotency keys protect backend business operations but do not prevent duplicate physical printouts.