Content
80%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is an exceptionally lean, well-organized contract that appropriately pushes schema detail to a verified one-level-deep reference. Its main weakness is workflow clarity: the author-to-build-to-verify sequence and the repair feedback loop are implicit in the body and only materialize in the reference, leaving validation checkpoints to be inferred rather than followed.
Suggestions
Add a short numbered authoring sequence to the body (draft summary/layout_policy, build each slide's layout_tree, run the task-local builder, verify against the contract) so the end-to-end workflow does not have to be reconstructed from the reference.
Surface the verification loop inline in the Build contract: state that after building, actual object bounds are compared with the contract and out-of-bounds or overlapping objects are repaired per the reference's repair order.
Inline a minimal one-slide JSON snippet (or enumerate the allowed `classification` values) so the required contract is executable without first opening the reference.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~26-line body is dense with contract specifics and contains no filler, no explanations of concepts Claude already knows, and no padding; every sentence adds constraints (e.g. 'No renderer may decide placement, shrink text, or infer layout after the audit.'). Nothing could be trimmed, so anchor 4's 'minor instances of over-explanation' does not apply. | 5 / 5 |
Actionability | Guidance is highly concrete — required per-object fields, named native kinds, a 9 pt minimum, `Inches(...)` bbox mapping, and rejecting zero or negative geometry before adding an object — but there is no inline example of the JSON shape, and terms like `classification` values and `layout_design` objects are only illustrated in the reference. Mostly executable with minor gaps matches anchor 4; anchor 5 would require inline common-case examples covering the schema. | 4 / 5 |
Workflow Clarity | Sections are logically ordered (contract, authoring rules, build contract) and one inline checkpoint exists ('Reject zero or negative geometry before adding an object'), but the end-to-end author, build, verify, repair sequence is implicit in the body — the rebuild-and-compare feedback loop lives only in references/layout-contract.md. Steps listed with checkpoints missing or implicit matches anchor 3; anchor 4 would require the sequence with most checkpoints present in the body itself. | 3 / 5 |
Progressive Disclosure | The under-50-line body is cleanly sectioned and schema detail is appropriately split into a single, real, one-level-deep reference, clearly signaled ('See `references/layout-contract.md` for the compact schema and repair guidance') and verified to exist with exactly the promised compact schema and repair guidance. This matches the clear-overview, well-signaled-reference anchor. | 5 / 5 |
Total | 17 / 20 Passed |