For agents

A coding agent's training data predates Lattice, so it has no working knowledge of it. Ask one to edit a lattice.json and it guesses from the config files it has seen before. This skill gives it the real commands, fields, and error messages instead.

Install the skill

Run it in the repo you want the skill in. The CLI finds the coding agents you have installed and writes the skill where each of them looks for it. It records what it wrote inskills-lock.json.

Load it for one session instead

This prints the skill as a prompt and writes nothing to disk. Use it for a one-off task, or for an agent the installer doesn't know about.

What's in the skill

One instruction file, plus four references the agent reads only when a task needs them. The list below is the whole skill.

SKILL.mdHow to orient in a repo that uses Lattice, run and preview tasks, declare a workspace, and decide what to cache.

references/configuration.mdEvery lattice.json field, its type, its default, and the error it raises when it is wrong.

references/cli.mdEvery command and flag, the exit codes, the environment variables, and which of them wins.

references/toolchains.mdThe built-in driver table, the engines that can be version-checked by name, and how provisioning works.

references/troubleshooting.mdSymptom, cause, and fix for the failures a run produces.

The source lives in the repo atskills/lattice/SKILL.md. Read it before you install it. It is plain Markdown, and it is the same text your agent acts on.

Without the skill

An agent that can read files and run commands gets most of the way on its own. Point it atlattice.json, then have it run lattice run <task> --dry-run. That validates the config and prints every resolved command without running one. For flags,lattice --help and each subcommand's --help are authoritative. Theconfiguration andCLI pages cover the same ground, and every page on this site is Markdown in the repo.