Skip to content
Poisson Labs Docs
GitHub poissonlabs.ai

ProveCommands

`poisson-prove init`

Writes a sweep spec and nothing else: the keyless example, or a commented template to edit for your own rollout. With --agent it writes instead the setup your own coding agent follows. It asks no question and never touches a store.

Terminal window
poisson-prove init --help
Output38 lines
usage: poisson-prove init [-h] [--example] [--target pkg.module:rollout]
[--out PATH] [--force] [--agent] [--repo DIR]
[--print] [--json] [--store PATH] [--plain]
[--debug]
Write a sweep spec, ./prove.toml unless --out names another file, and nothing
else: a commented template to edit, or with --example the keyless quickstart,
runnable as it is. With --agent, write instead the setup your own coding agent
follows (a skill and the prompt to give it), in the two folders coding agents read
skills from. Nothing is asked.
options:
-h, --help show this help message and exit
--example write the keyless quickstart, which runs as it is
(default: the template)
--target pkg.module:rollout
fill the template's target with this rollout (default:
a commented placeholder)
--out PATH the file to write (default: ./prove.toml)
--force replace the file if it exists (default: refuse)
--agent write the setup your own coding agent follows, a skill
and its prompt, instead of a sweep spec
--repo DIR with --agent: the repository to write the setup into
(default: this folder)
--print with --agent: print the skill and write nothing
(default: write the files)
--json print the file written as one JSON document instead of
text
--store PATH the store folder (default: $POISSON_PROVE_STORE when
set, else ./prove-runs)
--plain print plain text, as a pipe gets it, on a terminal too
(default: pretty on a terminal, plain otherwise)
--debug also print the traceback of an unexpected failure
examples:
poisson-prove init --example
poisson-prove init --target my_agent:rollout
poisson-prove init --agent
  • --example writes the keyless quickstart: the mock agent over three tasks and three tool failure rates, 194 replicates each. It runs as it is: poisson-prove sweep next.
  • --target pkg.module:rollout fills the template’s target with your rollout, so sweep finds it without --target.
  • --out PATH writes to another file than ./prove.toml; sweep then takes that file’s name.
  • --force replaces a file that exists. Without it an existing file is refused and left as it is.
  • --agent writes the setup for your coding agent instead of a spec: a skill, SKILL.md, and the prompt to give it, prompt.txt, in .agents/skills/poisson-prove-setup/ and .claude/skills/poisson-prove-setup/. Set Prove up with your coding agent says what the skill asks of your agent.
  • --repo DIR, with --agent, writes the setup into another folder.
  • --print, with --agent, prints the skill and writes nothing.
  • --json prints what was written as one JSON document.
  • --store PATH, --plain and --debug are the options every verb takes: the store folder (init reads none), plain text on a terminal too, and the traceback of an unexpected failure.

The keyless example, then a template for your own rollout in a second file:

Terminal window
poisson-prove init --example
Output2 lines
ok wrote prove.toml: a sweep of the keyless mock agent, 9 conditions x 194 replicates
next: poisson-prove sweep
poisson-prove init --target my_rollout:rollout --out mine.toml
Output2 lines
ok wrote mine.toml: a template; set its axes to what you vary
next: poisson-prove sweep mine.toml

A file that exists is refused, and the refusal says how to replace it:

Terminal window
poisson-prove init --example # exits 1
Output2 lines
error: prove.toml exists; init replaces it only with --force
next: poisson-prove init --example --force

The setup for your coding agent, and the same as JSON for a script:

Terminal window
poisson-prove init --agent
Output7 lines
ok wrote the setup for your coding agent (4 files)
wrote .agents/skills/poisson-prove-setup/SKILL.md
wrote .agents/skills/poisson-prove-setup/prompt.txt
wrote .claude/skills/poisson-prove-setup/SKILL.md
wrote .claude/skills/poisson-prove-setup/prompt.txt
prompt Use the poisson-prove-setup skill in this repository: set Poisson Prove up for me to review, then stop.
next: give your coding agent the prompt above, then review what it wrote
poisson-prove init --example --out other.toml --json
Output7 lines
{
"schema": "poisson-prove.init/1",
"verdict": "ok",
"wrote": ["other.toml"],
"next": ["poisson-prove sweep other.toml"]
…
}
  • 0: the file, or the four files of --agent, were written (or printed, with --print).
  • 1: refused, and nothing was written: a file exists and --force was not given, or a folder cannot be written.
  • 2: the command line is wrong, such as --repo without --agent.
  • 3: an unexpected failure; --debug prints its traceback.
  • 130: interrupted with Ctrl-C.

./prove.toml, or the file --out names; with --agent, the four files above in the folder, or in --repo. Nothing else: no store, no .gitignore entry, none of your agent’s own files (AGENTS.md, CLAUDE.md, .cursor/rules/).