Content
50%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 well-sectioned and contains a useful concrete vocabulary table and prompt pattern, but its core workflow defers to reference files that are absent from this bundle, and much of the Output Contract reads as abstract policy rather than executable guidance. Tightening the prose and either shipping or inlining the deferred vocabulary would materially improve it.
Suggestions
Ship the referenced files (references/directors-read.md, references/vocab/ru.md, references/surface-prompt-profiles.md) in the bundle or inline the essential vocabulary, since every 'load X' instruction currently points at a missing file.
Rewrite the Output Contract as a short numbered checklist of concrete actions (what to return, what to preserve, what to ask), replacing abstract statements like "distinguish their visible performance, staging or information tradeoffs".
Remove duplication and meta-commentary: the token-does-not-attach-file rule appears twice, and the Intent section's remarks about the empty review artifact do not guide execution.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and vocabulary tables earn their tokens, but there is avoidable padding: the Intent section is meta-commentary ("The shipped independent review artifact is empty, so treat these choices as working production wording pending locale-specialist review") and the idea "A written example token does not attach a file" / "A written token does not attach an asset" is stated twice. It matches 'mostly efficient but could be tightened', not anchor 4's 'only minor instances'. | 3 / 5 |
Actionability | There are genuinely concrete elements — the Russian wording table ("медленный наезд камеры", "контровой свет"), the copy-paste Compact Pattern, and the named slop words ("кинематографичный", "эпичный") — but the core workflow depends on files ("load [vocab/ru](../../references/vocab/ru.md)", "load the [Director's Read](../../references/directors-read.md)") that do not exist in this bundle, and much of the Output Contract is abstract ("distinguish their visible performance, staging or information tradeoffs") rather than executable guidance. This lands between anchors 3 and 4, closer to 'incomplete; missing key details'. | 3 / 5 |
Workflow Clarity | An opening sequence exists (load Director's Read → classify the brief → complete the record → translate into carriers) but it is prose, not a numbered workflow, and validation checkpoints are only implicit ("Time the actual performance", "name the criterion, propose one concrete change"). Anchor 3 ('sequence present but checkpoints missing or implicit') fits; anchor 4 would require explicit checkpoints tied to each step. | 3 / 5 |
Progressive Disclosure | References are clearly signaled with purpose (Director's Read for classification, vocab/ru for dense vocabulary, surface-prompt-profiles for example adaptation), but the bundle contains no references/ directory — every referenced path (../../references/directors-read.md, ../../references/vocab/ru.md, ../../references/surface-prompt-profiles.md) resolves outside the skill or to nothing, so navigation is broken and the vocabulary payload the body defers to is unavailable. Anchor 3 ('some structure; references present but... could be better organized') applies rather than 4's 'references mostly clear and appropriately placed'. | 3 / 5 |
Total | 12 / 20 Passed |