rekal-setuplisted
Install: claude install-skill rekal-dev/rekal-cli
# Rekal — get it running
This skill has one job: get Rekal installed and initialized. It does not answer
questions from memory. The binary installs a second, much richer skill that
does; after `rekal init`, route recall questions there, not here.
## What it is, in one paragraph
Every AI session settles things — why this approach, what got tried and thrown
away. Then the session ends and the reasoning is gone. Rekal captures each
session at commit time, stores it raw in git, indexes it locally, and shares it
across the team when the work merges. No server, no vector-database tier: the
store is two files in `.rekal/`. A later agent asks `rekal "<question>"` and
gets the conversation behind the change.
## Setup changes the user's machine and repository. Ask first — every step.
The install pipes a script from the internet into a shell. `rekal init` writes
git hooks, an orphan branch, and a line in `CLAUDE.md`. These are the user's
decisions, not yours. State plainly what each step does, then wait.
If the user declines, stop. Say Rekal is not available here and answer from the
tree. Declining is a complete answer, not a problem to route around.
## 1. Install the binary
Requires git, macOS or Linux. One binary — CLI, database, embedding model, all
embedded.
```bash
curl -fsSL https://raw.githubusercontent.com/rekal-dev/rekal-cli/main/scripts/install.sh | bash
```
Installs to `~/.local/bin`; `--target <dir>` overrides. Verify with `rekal
version`.
If the shell still repor