skill

name	skill
description	mokata · Author a new mokata skill test-first (RED-GREEN-for-docs); the write is human-gated.
argument-hint	<name> --summary <s> --require <doc>:<must-contain> --content-file <f>
allowed-tools	Bash, Read, Write

mokata · skill (author your own governed command)

Author a new /<name> skill the mokata way: test-first for docs (G6). You declare what the skill's content MUST contain (--require name:must-contain, repeatable); the draft is checked against those requirements (RED) and only a passing draft (GREEN) is promoted to a command. Writing the rendered command is a durable write, so it is human-gated.

How to run

1. Draft the content to a markdown file (use Write), then decide the doc requirements it must satisfy — each as --require <name>:<must-contain>.

2. Resolve the bundled engine (read ~/.mokata/plugin-root → ROOT, or a mokata CLI on PATH):

   PY="$(command -v python3 || command -v python)"
   ENGINE="PYTHONPATH=\"$ROOT/src\" \"$PY\" -m mokata"
   

3. Check it RED→GREEN first (writes nothing while RED):

   eval "$ENGINE skill author <name> --summary \"mokata · <one-liner>\" \
     --require <doc>:<must-contain> --content-file <draft.md> --path ."
   

   If it reports RED (a requirement unmet), revise the draft until every requirement passes.

4. Approve the write — only after the draft is GREEN and the user confirms, add --yes:

   eval "$ENGINE skill author <name> --summary \"mokata · <one-liner>\" \
     --require <doc>:<must-contain> --content-file <draft.md> --yes --path ."
   

   The write goes through the universal gate (a secret in the content is hard-blocked even when approved). Report where the new skill was written.