skill-creator-linter

Run the lint script with --skill-md pointing at the file or pipe it via --skill-md-stdin, passing --gates G1,G2. The script prints JSON with passed flags and diagnostics for each gate.

G1 parses the plan with parse_meta_plan, verifies every agent or skill_exec step references an existing catalog skill, rejects nested meta-skill references, and greps for user_message templates missing the xml_escape or slugify filter.

G2 replaces step executors with stubs that yield a done event, then runs the DAG scheduler over the parsed plan. It passes if no exception is raised and the topology terminates, reporting the number of steps visited.

The runtime agent executor refuses nested meta-skills with a 'cannot compose another meta-skill' error. The linter rejects them at G1 so candidates do not pass validation only to crash at runtime.

No. It is marked user-invocable false and disable-model-invocation true, so it runs only as a DAG step called by meta-skill-creator, though the script can be executed manually from a shell.