authoring_syntaxlisted

# authoring_syntax — write for the shell that loads it House syntax for text authored for AI consumption — skill bodies, shell focus/mandate text, agent prompts. Governs how the text reads; lifecycle mechanics (create / seed / grant / persist) stay in `local_skill_management`. Retrofit = touch-time: an existing doc converts when next edited. No mass pass. Engine skill bodies are upstream-owned — a syntax fix to one is an upstream PR (`issue_reporting` boundary), never a fork edit. ## Fires when - authoring / editing any AI-consumed doc - "syntax check X" -> run Self-test against X, return findings ranked, no rewrite ## Rules - directive + success condition. every instruction = action + observable pass. "bounce the api -> `make health` returns ok" not "make sure the api is running". - prohibitions are specific. bar the exact action: "NEVER `echo $SECRET`" not "be careful with secrets". no compliance test = cut or sharpen. - operators, light: `->` then / leads-to. `=` is / defined-as. `/` or. `+` and. use only where compression loses nothing; else prose. - need-to-know. cut the why unless it changes the action taken. NEVER name a tool / path / failure mode the reader would not otherwise encounter. - CAPS = emphasis. rare — caps on every rule = caps on none. - imperative voice. "validate input" not "you should validate input". - format: prose = reasoning. tables / short bullets = structured data. `path:line` = code. - one source of truth. state a constraint o