Content
96%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 dense, highly operational playbook: fully executable tool-level guidance, an 11-step sequence with explicit approval gates and human-confirmation feedback loops, and disciplined offloading of detail to references — with no token spent teaching Claude things it already knows. The only structural weakness is bundle hygiene: one orphaned reference file and atom paths that don't resolve inside the bundle as shipped.
Suggestions
Reference references/excellent-script-example.md from the script step (Step 5 / script-craft.md) or remove it from the bundle — as shipped it is orphaned and cannot be discovered through navigation.
Make the atoms/ paths resolvable or annotate where they live (e.g. a one-line note that atoms are shared across the marketing skills), since atoms/write-script/ATOM.md and its siblings do not exist inside this bundle.
Consider a short 'Bundle map' line at the top (one line per references/ file) so the reference set, including currently unreferenced files, is visible at a glance.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is relentlessly operational — tool calls, file paths, gate conditions, exact sentences to say — with zero explanation of concepts Claude already knows (no 'what a video ad is', no library or platform primers). Every clause carries domain-specific information Claude could not infer (e.g. 'expect candidates_too_similar if two say the same thing', 'budget_exceeded → stop and ask'), matching the 5 anchor 'every token earns its place'; a 4 would require over-explanation that could be trimmed, which is absent. | 5 / 5 |
Actionability | Guidance is fully executable for an instruction-only skill: exact tool calls with parameters ('gooseworks:video_batch more_like (change: "hook", free)', 'video_read (include: ["renders","recipe","scenes"])'), exact phrasing for the human ('Connect GooseWorks first — I make your video ads through it'), file paths ('ads/videos/<project_id>.md'), and error branches ('budget_exceeded → stop and ask'). This matches the 5 anchor 'copy-paste ready… specific examples cover the common cases' — chat vs shell hosts, batches, and rejection fixes are all covered. | 5 / 5 |
Workflow Clarity | Steps 1–11 are clearly sequenced with explicit validation checkpoints and feedback loops: script-check atoms run on every candidate 'before saving', the gate ('show it, price it, wait') blocks rendering until video_approve records the marketer's exact words, Step 8's honest review labels findings 'please confirm' with 'Two rounds at most', and budget_exceeded halts production. This matches the 5 anchor (explicit validation, error-recovery loops, checklists for complex processes); spend-bearing operations all pass through gates, so no cap applies. | 5 / 5 |
Progressive Disclosure | Structure is good: per-step load signals ('Load references/reading-the-brain.md', 'references/paths-and-gates.md §3') point to one-level-deep files that all exist, keeping the overview lean. Two minor gaps keep it from the 5 anchor: references/excellent-script-example.md is present in the bundle but referenced nowhere in SKILL.md or the other reference files (orphaned, undiscoverable), and the four atoms/*/ATOM.md paths do not resolve within this bundle. This fits the 4 anchor ('most content appropriately placed; references mostly clear; minor organization gaps') better than 5, and is well above 3 ('references present but not clearly signaled'). | 4 / 5 |
Total | 19 / 20 Passed |