New harness
Stamp a harness beside your work — the default layout places it under harness/. --mode new skips the placement question; --name and --out answer the rest.
Setup · Dry-run first
Set up Paniolo around
the repositories you
already have.
Start a new harness, adopt this repo, or connect it to one you keep elsewhere. A short guided interview asks what it needs — every answer is also a flag, so anything interactive replays as one command. Run a dry run first; nothing changes until you apply.
Runs locally. --dry-run changes nothing.
npx @paniolo/cli init --dry-run
New, adopt,
or connect.
Adopt this repo
Adapt Paniolo to guidance that is already there — retrieval hooks and MCP registration are added to your existing files, marked, and removable. --no-retrieval keeps only the safety hooks.
Connect
Wire this repository to a harness you keep elsewhere with --mode connect --connect-to ../harness — the relationship is named once in paniolo.config.json.
No flags also works: init detects what it can — existing harness, sibling
candidates, config — and asks the placement question it can't answer. It
prints the underlying evolve harness command before it runs
it, so a guided run teaches the commands it wraps.
Keep what
works.
Paniolo adds marked sections to your instruction files instead of
replacing them, and one paniolo.config.json names each
repository once — a config you can read.
Give your agents a knowledge base
--wiki stamps an agentic knowledge base beside the
harness, already wired so supported agents read it on every prompt.
--no-wiki skips it — add one later with
paniolo wiki init.
Interview once, replay anywhere
The interview ships today: --print-answers emits the
resolved intent as versioned JSON, and
--answers answers.json replays it non-interactively —
the same setup in CI that you walked through locally.
Every question has a flag
No prompt-only answers: --mode, --name,
--out, --wiki/--no-wiki,
--repo name=path:kind, --tool,
--yes. Without a terminal, a missing answer is an error
naming the flag — never a guess.
If you're
hesitating.
We already wrote instructions.
Init adds marked sections to the files you have and leaves the rest alone. Adoption is designed to be removable — your content stays yours.
We already keep docs somewhere.
Docs describe the code. The knowledge base holds what the code can't — why decisions were made, what you rejected, what you learned — where agents read it automatically on each prompt.
We do not want a knowledge base yet.
Skip it with
--no-wiki and add it later with paniolo wiki init.We already have a config.
paniolo evolve config migrate previews the upgrade before writing — and evolve config effective shows what every repo inherits.