What problem does it solve? When a CLI tool gains a second output format (for example a stdlib-only HTML emitter beside a python-pptx assembler), ad-hoc implementations fork the data model, duplicate stamp logic, and corrupt prior outputs on failed writes. This Skill enforces the architectural discipline that keeps multiple format assemblers consistent over one shared intermediate representation. ## Core Features & Use Cases - Shared IR consumption: Requires the new assembler to read the existing format-agnostic intermediate representation, extending the IR itself rather than creating a format-specific side channel. - Single provenance builder: Routes every assembler's stamp (source path plus sha256 content hash) through one shared builder so implementations cannot drift apart. - Explicit format discrimination and atomic writes: Mandates a named format field on the shared result dataclass and atomic output via temporary-file-then-os.replace, so interrupted writes never destroy prior good output. - Use Case: Adding a zero-dependency HTML deck assembler next to an existing python-pptx one, with disclosure and exit codes keyed by (item, format) so a partial success in one format is never reported as a full success or failure. ## Quick Start Add a second output-format assembler over the existing shared IR in this CLI tool, following the format-assembler-discipline rules for stamps, atomic writes, and per-format results.
