Content
81%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 highly actionable, well-validated operational guide whose tool documentation and workflow gating are exemplary. Its weaknesses are token efficiency — the same safety rules are restated many times across sections — and some deep policy detail that belongs in the already-existing video reference rather than SKILL.md.
Suggestions
Consolidate repeatedly stated rules (first-Content-is-main, wait/timeout-is-not-failure, retry gating) into a single authoritative section and point to it from the tool sections instead of restating them, cutting significant length without losing enforcement.
Move the audioPolicy edge-case detail (native_mute/mute_fallback, composition vs direct-generation field shapes) and the final-video composition deep dive into references/video-generation.md, which video tasks are already required to load.
Merge the six language/terminology bullets in Ground rules into two: one on consistent user-facing language with the term-pair table, and one on keeping identifiers/paths unchanged.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense domain policy with no filler or explanation of known concepts, but key rules are systematically duplicated — 'the first Content is automatically main' appears ~6 times (Ground rules, Capabilities, Confirmation, add_canvas_contents, select_canvas_main_content, Workflow coverage) and timeout/retry semantics ~5 times. This goes beyond anchor 4's 'minor instances that could be trimmed' while stopping short of anchor 2's padded sections. | 3 / 5 |
Actionability | Every one of the 13 tools gets an executable, copy-paste-ready snippet with the required import shown once, parameter tables with required/optional flags, exact field names, and realistic example values (e.g. the add_canvas_elements items payload and the design-marker reference_images example). Fully matches anchor 5. | 5 / 5 |
Workflow Clarity | Multi-step flows are explicitly sequenced (7-step media-transformation pattern, 4-step marker processing) with hard validation checkpoints: the mandatory preflight before every write, re-query after writes, staged user-confirmation gates, bounded retry gated on explicit status=failed with a 2-attempt cap, and post-import verification via query_canvas_element_details. This is the anchor-5 pattern of sequence + validation + feedback loops. | 5 / 5 |
Progressive Disclosure | All 9 referenced files exist, and the Reference routing section plus Tool map clearly signal one-level-deep conditional loading. However, the ~600-line main file inlines deep detail that overlaps existing references — notably the ~250-word audioPolicy edge-case paragraph and the final-composition section, which duplicate territory of references/video-generation.md — a minor organization gap that keeps it at anchor 4 rather than 5. | 4 / 5 |
Total | 17 / 20 Passed |