Blog

Guides and research on GitHub Actions, CI/CD, and AI code review
from the team building Tenki's runners, code reviewer, and agent sandboxes.

The Harbor and Tenki logos side by side, joined by a plus sign, over a dark blue isometric cube field

Sandbox

How to benchmark inference providers across coding-agent harnesses

Test any model or inference endpoint across coding-agent harnesses like Claude Code and OpenCode on Terminal-Bench, with every trial in a Tenki sandbox.

Oct 5, 20266 min read

The Omnara and Tenki logos side by side, joined by a plus sign, over a dark blue isometric cube field

Sandbox

Omnara agents can now run on Tenki Sandboxes

Omnara now supports Tenki Sandboxes as a machine pool provider. Add a Tenki API key and your Omnara agents get their own Linux VMs on demand.

Oct 1, 20264 min read

The Naïve and Tenki logos joined by a plus sign, over a dark blue isometric cube field

Sandbox

Naïve agent computers now run on Tenki Sandboxes

Naïve's AI employees incorporate companies, launch campaigns and write code. Every task they run now gets its own fresh Tenki sandbox.

Sep 30, 20262 min read

Sandbox

Keep API tokens out of your sandbox

Tenki swaps a placeholder for your real API token outside the sandbox, so agent code can call an API without ever holding the credential.

Sep 30, 202610 min read

Isometric illustration of glowing blue disk stacks slotting into a row of runner blocks on a dark grid

CI/CD

Sticky disks keep your GitHub Actions caches warm between jobs

Tenki sticky disks keep dependency and compiler caches on a disk that follows the cache key, so jobs start warm without restoring a tarball.

Sep 29, 20267 min read

The Anchor and Tenki logos joined by a plus sign, over a dark blue isometric cube field

Sandbox

Connect a sandboxed AI agent to Anchor Browser over CDP

Run your agent in a Tenki Sandbox and drive an Anchor-managed browser with Playwright, with the API key kept out of images, templates and snapshots.

Sep 22, 20268 min read

Abstract chart of four bars of filled squares measured against a marked vertical line

CI/CD

How to benchmark GitHub Actions runners fairly

Compare CI runners on the same YAML, warm caches and repeated runs, measure cost per job instead of rate per minute, and see what our benchmark shows.

Sep 22, 202610 min read

Abstract stack of browser windows, the front one enclosed by a dashed boundary

Sandbox

A browser for AI agents: run Chrome in a disposable sandbox

Give an AI agent or computer-use model its own Chrome in a throwaway VM, driven over CDP with Playwright or Puppeteer and watchable through noVNC.

Sep 22, 20268 min read

Sandbox

Run your tests on a warm sandbox before you push

Sync your uncommitted working tree to a warm Linux sandbox with Crabbox and run your test suite there, so your laptop and your coding agent stay free.

Sep 22, 20269 min read

Abstract grid of ten outlined cells with a single cell filled solid blue

Sandbox

Where to run code your AI agent wrote

Your laptop, a container, a CI runner or a disposable VM? Where to run agent-generated code and untrusted MCP servers so a bad result stays contained.

Sep 22, 20269 min read

Abstract diagram of three panels, each with more of its slots filled than the last

CI/CD

Sign and notarize macOS and iOS builds in GitHub Actions

Import a .p12 into a temporary keychain, install profiles, notarize with notarytool or use fastlane match, without leaving signing keys on the runner.

Sep 22, 20269 min read

Abstract diagram of a central highlighted config panel linked to four smaller panels

AI Code Review

AGENTS.md and CLAUDE.md: shared instructions for AI coding agents

How Claude Code, Codex, Copilot and Cursor read AGENTS.md and CLAUDE.md, what to put in the file, and the anti-patterns that make it useless.

Sep 4, 20269 min read

Abstract diagram of a rising line crossing a dashed boundary over a panel of stacked rows

CI/CD

GitHub Actions cost optimization: how to cut your CI bill

Reduce GitHub Actions costs with runner right-sizing, caching, concurrency controls, workflow audits, spend alerts and a lower per-minute rate.

Aug 5, 202614 min read

Abstract diagram of one blue block linked to a bounded grid of dimmer cells

CI/CD

Self-hosted GitHub Actions runners: ARC, security, and cost

How self-hosted GitHub Actions runners register and autoscale with ARC, an 8-point hardening checklist, and the operational work you take on.

Jul 31, 202612 min read

Isometric illustration of a card on a conveyor between two glass meters, one holding coin stacks and one a runner cube, both feeding one stack of sheets

CI/CD

Copilot code review cost: AI credits plus Actions minutes

Copilot code review on private repos bills AI credits and GitHub Actions minutes. What each review costs, how to measure it, and which runners it can use.

May 22, 20268 min read

Abstract grid with a fully highlighted top row above rows of scattered hits

AI Code Review

How we benchmark AI code reviewers: per-bug scoring on real PRs

How Tenki's code review benchmark scores seven AI reviewers bug by bug on 50 real pull requests, and what per-bug scoring does and doesn't show.

May 19, 20267 min read

Abstract diagram of nested rectangles growing brighter toward a solid blue block

CI/CD

The hidden CI tax of AI coding agents

AI coding agents open more PRs and push more often, and every push runs CI. How to model the extra cost, measure the agent share, and cut it.

Apr 8, 202610 min read

Isometric illustration of a small robot cube holding up a code diff panel, a magnifying lens over one red line and a ticked checklist beside it

AI Code Review

Reviewing AI-generated code: a checklist for agent-written PRs

Seven ways code from AI coding agents goes wrong while looking right, with an example of each and what to check before you approve the PR.

Mar 30, 202612 min read

Isometric illustration of workflow sheets on a conveyor under a magnifying lens, with two lines flagged red and checked sheets past it

CI/CD

Lint GitHub Actions workflows with actionlint: setup, config and CI

Install actionlint, read its errors, set custom runner labels in .github/actionlint.yaml, and run it in CI with problem matchers, reviewdog or pre-commit.

Mar 28, 202612 min read

Abstract dependency graph in which three connected nodes are highlighted

CI/CD

Monorepo CI with GitHub Actions: selective builds done right

Run only the CI jobs a change affects, using path filters, Turborepo and Nx affected builds, dynamic matrices, required checks and runner sizing.

Mar 6, 202617 min read

Isometric illustration of a frame holding a chain of connected job blocks, with small step tiles moving from an open toolbox into one block

CI/CD

Reusable workflows vs composite actions in GitHub Actions

When to share GitHub Actions config as a reusable workflow or a composite action, covering secrets, runners, logs, nesting limits and versioning.

Feb 6, 202614 min read

Isometric illustration of three glass cache cylinders wired into a row of runner blocks, with package boxes and containers on a conveyor beside them

CI/CD

GitHub Actions caching strategy: dependencies, builds and Docker

How the GitHub Actions cache matches keys, scopes and evicts entries, and how to cache npm, pnpm, Go, Python, Rust, Turborepo and Docker layers.

Jan 16, 202616 min read