KATA is Cerb's configuration language: keys, values, and indentation. It looks like YAML and is not YAML. Everything below applies everywhere Cerb uses KATA -- automation scripts and policies, toolbars, sheets, charts, workflows, interaction elements, agent config, the schema file -- not just automations.
\nFull reference: /cerb-docs/references/docs/kata.md.
A bare key: with anything indented under it is an OBJECT. Always. There is no exception and no heuristic.
To put a multi-line VALUE under a key, the key must carry an annotation. @text: is the generic one, but any annotation works -- @int:, @bool:, @json: all accept a value indented on the following lines.
message@text:\n Subject: Order 123\n Status: open\n→ the string \"Subject: Order 123\\nStatus: open\".
Drop the annotation and the same lines mean something completely different:
\nmessage:\n Subject: Order 123\n Status: open\n→ the object {\"message\": {\"Subject\": \"Order 123\", \"Status\": \"open\"}}.
That second one does not error. It parses cleanly into the wrong shape, and the failure surfaces much later as an empty placeholder or a command that received an object where it wanted a string. Prose is especially prone to it, because a colon anywhere in a sentence is enough -- Subject:, Note:, a time like 9:00.
You are only warned when the indented text can't be read as keys:
\nnote:\n See https://cerb.ai/docs/ for more\n→ Unexpected syntax (line 2). Same mistake, but this one is loud. Don't rely on it.
So: any value that spans lines, or that contains prose, needs an annotation on its key. When in doubt, @text:.
Values are literal text from the colon to the end of the line. Quotes are not syntax -- they become part of the value.
\nname: Jeff → \"Jeff\"\nname: \"Jeff\" → \"\\\"Jeff\\\"\" ← the quotes are IN the string\nname: 'Jeff' → \"'Jeff'\"\nNever wrap a value in quotes to \"make it a string\". It already is one. Quote only when the quotes are genuinely part of the text you want.
\nA # only starts a comment at the beginning of a line (after indentation). There are no trailing comments.
# This is a comment.\nname: Jeff\nname: Jeff # this is NOT a comment\n→ the value is \"Jeff # this is NOT a comment\".
Note that this cuts the other way inside an annotated text block: there, a # line is content, because the whole block is a literal value.
All values are text by default. Convert explicitly:
\n@bool @int @float @csv @json @list @date @key @text @raw @optional
Annotations combine with a comma (@optional,text:). @optional is honored by whatever CONSUMES the KATA rather than by the parser, so what it does depends on the command -- check that command's documentation rather than assuming it drops empty values.
Never stack @raw inside an already-raw block. Inside script@raw: / event_kata@raw: / policy_kata@raw:, use the normal annotations (if@bool:, set:). Writing if@raw,bool: {{...}} makes the condition never match.
A repeated key silently clobbers the earlier one. Where you need the same command twice under one parent, give each an alias with /:
set/init:\nset/increment:\nawait/step:\nThe part before the / is what Cerb dispatches on; the part after is yours, and it may contain only letters, numbers, and underscores -- no dots.
&name: defines a reusable block; @ref uses it. Useful for a long value you'd otherwise repeat, and for keeping a big text block out of the middle of a script.
Two-space indentation, spaces not tabs. Annotate types rather than relying on a reader to infer them.
\n","files":[{"path":"SKILL.md","sizeBytes":4161,"isText":true}],"reviewScore":null,"reviewSummary":null,"trust":{"provenance":"trusted-source-unreviewed","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow.","bodySource":null},"bodyLocked":false,"purchaseUrl":null,"sourceUrl":null,"report":{"provenance":"trusted-source-unreviewed","screen":{"ran":true,"outcome":"clean","suspicious":0,"notes":0,"hiddenCharacters":false},"virusScan":{"engine":"clamav","status":"clean","scannedAt":"2026-10-16T16:28:15.064518Z","sha256":"DED62BB63989456920FFECD9D328D61B81719D30FCF28A5BD6CD530188CF7C77","sizeBytes":2074},"review":null,"source":{"repositoryUrl":"https://github.com/jstanden/cerb","path":"features/cerberusweb.core/assets/agent_filesystems/cerb-agents/skills/kata","license":null,"commit":"cf28b62ee0f67ff993645931568a4e4bbad074f0","subtreeSha":"AADC5B463D2813524A800BBDA0ECC7B64D4FA2C5E9B6299F4D9185A5CD9509DB","lastSyncedAt":"2026-10-16T16:27:56.786317Z"},"reviewedAt":"2026-10-16T16:28:43.147999Z","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow."},"install":[{"target":"skills-cli","command":"npx skills add https://github.com/jstanden/cerb/tree/12.0.x/features/cerberusweb.core/assets/agent_filesystems/cerb-agents/skills/kata"},{"target":"claude-code","command":"claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install jstanden-cerb@llmmart"},{"target":"git","command":"git clone https://github.com/jstanden/cerb.git"}]}