Content
71%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.
A tight, example-driven policy skill whose blocking and action-hierarchy sections are genuinely actionable, weakened mainly by a dangling output format and by references that resolve outside the bundle to files that are not present. The rhetorical Intent section is the only noticeable token waste.
Suggestions
Include one worked example of the mandated output — a filled-in character card with tag map, tier assignments, and continuity constraints — so the Output Contract is unambiguous.
Fix or inline the `../../references/shot-list-continuity.md` and `directing-engine.md` links: they escape the skill bundle and the target files are absent, so either bundle them under `references/` with relative paths or move their essential content inline.
Compress the Intent section to a single operational sentence (e.g., "Treat every recurring character as a continuing cast member with a contract, never a re-cast stranger") and add a one-line final self-check that every visible character has exactly one action tier and no ambiguous pronouns remain.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and prescriptive — a field table, tiered rules, and concrete example prompts ("Character A lowers the envelope; Character B remains in the doorway") with no explanation of concepts Claude already knows. Not 5: the Intent section ("The user is protecting someone here... the user feels it as betrayal, not as a rendering artifact") is rhetorical framing rather than operational guidance and could be trimmed to one line. | 4 / 5 |
Actionability | For an instruction-only skill the guidance is unusually concrete: stable tag formats (`Character A`, `@Image1 subject`), a field-by-field prompt table, an explicit three-tier action hierarchy with example phrasing ("lip corner lifts and she holds a half-second glance"), and explicit prohibitions ("explicitly exclude standing, walking, turning"). Not 5: the Output Contract mandates "a character card, tag map, action assignments" but no example of that output's shape is ever shown, leaving the deliverable underspecified. | 4 / 5 |
Workflow Clarity | The sections read as a coherent implicit sequence — assign tags → block multi-character actions → apply the tier hierarchy → hands/face and likeness constraints → inherit sequence state → emit the output contract — and the Output Contract gives a verifiable end product. Not 5: nothing tells Claude to self-check (e.g., that every visible person received exactly one tier, or that no ambiguous pronouns remain after more than one character appears), so checkpoints are implicit rather than explicit; not 3 because the flow is coherent and no destructive/batch operation is involved. | 4 / 5 |
Progressive Disclosure | Sections are well organized for a short skill, and the two outbound references are clearly signaled with load conditions ("Load shot-list-continuity when... identity, wardrobe, props, eyeline... must survive across multiple shots"). However, both links point to `../../references/` — outside the skill bundle — and those files do not exist in the provided structure, so navigation cannot be verified and the links may be broken; this is more than the "minor organization gaps" of anchor 4. | 3 / 5 |
Total | 15 / 20 Passed |