Kiloforge Skills
16 skills that turn your terminal into an AI-powered development platform. Works with Claude Code, OpenCode, Amp, Codex, and Antigravity. The same skills used to build Kiloforge itself — from a blank repo to 80,000+ lines of code in 5 days.
Install
Requires Claude Code and git.
$curl -fsSL https://raw.githubusercontent.com/Kiloforge/kiloforge-skills/main/install.sh | sh
Become a Kiloforger
The skills are the same — how you use them changes as you scale. Start single-threaded to learn the process, then level up.
/kf-setup/kf-architect/kf-statusoptional/kf-developer/kf-setup/kf-architect/kf-status/kf-developerGetting Started
One terminal. One agent at a time. This is your training ground.
- 1.Install the skills (see above) and clone the repo you want to work on.
- 2.Start Claude Code with permissions bypassed so the agent can work uninterrupted:
claude --dangerously-skip-permissions - 3.Run
/kf-setupto initialize your project, then/kf-architectto create your first track. - 4.Hand the track to
/kf-developer <track-id>and watch the implementation unfold.
Building Confidence
Start here. Run a single agent at a time — architect a track, then hand it to a developer. Watch what comes back.
The goal is not to produce output at maximum speed. It is to understand the process and to build a calibrated sense of what agents produce given a well-scoped specification. When the architect writes a track with clear acceptance criteria and a phased plan, the developer follows it closely. You will find that the alignment between intent and implementation is higher than you expect. That confidence is the foundation everything else is built on.
As you build this intuition, pay attention to which kinds of work align well and which carry higher risk. Logic-heavy backend tasks — APIs, data models, business rules — tend to be highly predictable. Visual elements and UI work can drift further from expectations, especially when the spec describes aesthetics or layout nuance. Knowing where the risk sits lets you allocate your attention accordingly.
To mitigate it, write more specific acceptance criteria for visual work — reference existing components, specify spacing values, name the design tokens. Provide screenshots or describe the exact layout you want. The tighter the visual spec, the closer the output. For structural code, the spec can be looser because the constraints are more naturally enforced by types, tests, and APIs.
Work at the Decision Layer
At the rate you can instantiate new implementations, shipping code to some remote to run a CI process there makes no sense — ship it when you have time to ramp down and address issues en masse. The throughput can be far higher than any remote review pipeline can handle.
Your time is too valuable to spend at the wide end of the funnel, sifting through every change. Instead, work at the narrow end — where decisions have leverage. Determine which changes matter. Prune out the noise. Identify what actually delivers value to your users.
Kiloforge is not trying to be the best interface for chatting with a single agent — the CLI already does that well. It is designed to let you set individual agent sessions aside and focus on the high-level decisions that, presumably, only you can make.
Skills-only or full platform
These skills work standalone in any supported agent session — this is exactly how Kiloforge itself was built. For teams scaling to many parallel agents, the full Kiloforge platform adds a local control plane (Cortex), a real-time dashboard (Command Deck), and agent lifecycle management to handle the coordination problems that emerge at scale.