Command line

npx shibbo <command> [options]

Command	Does
check <files | dirs | ->	Score files and list hits with fixes
rules	Print the rule block for an instructions file
init <harness>	Set up rules, MCP and hooks for a harness
export <target>	Write the rule block into one file format
show <id>	One tell with its evidence
list	All tells; filter with --domain or --status
models	Measured models and release months
mcp	Start the MCP server on stdio
hook <event>	Hook entry point for harnesses (reads JSON on stdin)
--version	Print the installed version

check

shibbo check README.md docs/ --threshold 80
cat draft.md | shibbo check - --kind markdown

Option	Default	Does
--model <id>	$SHIBBO_MODEL	Weigh the phrases this model over-uses more heavily
--threshold <n>	0	Exit 1 when any file scores under n. A value that isn't a number exits 2
--format <f>	text	text, json or github (Actions annotations)
--all	off	Also print hits under their density limit, and the questions to judge by reading
--kind <k>	from extension	For stdin: markdown, text, html, css, code
--no-lexicon	off	Curated rules only, without measured phrases
--disable <ids>	none	Skip these tell ids, comma-separated
-q, --quiet	off	Print only files under the threshold

The file's extension picks the rules: prose tells for Markdown and text, prose and interface tells for HTML, interface tells for CSS, interface and code tells for JSX, TSX, Vue, Svelte and Astro, and code tells for other source files.

Folders are walked recursively, skipping node_modules, .git, dist, build and hidden folders. Exit status: 0 when every file meets the threshold, 1 when one doesn't, 2 on errors.

rules

shibbo rules --model gpt-6 --limit 10 --phrases 8

Option	Default	Does
--model <id>	none	Rank tells by this model's measured rates
--domain <list>	prose,chat	Which tells to rank: prose, chat, code, ui, agent
--limit <n>	14	How many rules
--phrases <n>	12	How many measured phrases to list, when a model is given

The block opens with a Say it clearly section and, unless --domain leaves out agent, closes its list with six agent habits.

init

shibbo init <harness> [--model <id>] [--dir <path>] [--dry-run]

init merges into existing config files, and hooks you already have stay. Harness ids: claude, codex, cursor, copilot, gemini, devin, cline, roo, aider, kiro, junie, amp, opencode, goose, zed, continue, warp. See the support table.

export

shibbo export <claude | agents | cursor | vale> [--model <id>] [--out <dir>]

claude, agents and cursor write a marked block that later runs replace. vale writes one Vale rule per pattern tell into styles/Shibbo/.

hook

shibbo hook <session-start | post-edit | stop> [--for <harness>]

Reads the harness's hook JSON on stdin and prints its feedback format. --for accepts claude (default), codex, cursor, gemini, copilot, devin, cline, kiro, junie. You rarely call this yourself: init writes the commands.