Content
88%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.
An unusually rigorous operational reference: concrete commands, exact tool parameters, and validation-heavy workflows with explicit error-recovery loops for batch uploads. Its main costs are a few overly dense paragraphs and heavy inline reference tables that could be split out.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Dense tables and terse, non-obvious platform rules (limits, dedup IDs, screening behavior) with no explanation of concepts Claude already knows, but a few overloaded paragraphs — e.g., the single-paragraph upload-status rules covering videoIds caps, failedChunks, and search caveats — could be tightened or split. Efficient overall with minor trimming opportunities, matching the 4 anchor rather than the every-token-earns-its-place 5. | 4 / 5 |
Actionability | Copy-paste ready commands (`roboflow import -w <workspace> -p <project-id> /path/to/dataset`, `project.version(1).download("yolov8")`) plus exact MCP tool names, parameters (`images_search(project_id, media_types=["video"])`), size limits, polling states, and concrete error strings fully specify the common upload, search, and versioning cases. | 5 / 5 |
Workflow Clarity | Multi-step flows are explicitly sequenced with validation checkpoints and feedback loops: the 5-step version pipeline, "poll `image_upload_status` until it is `completed` or `failed`", inspecting `uploadErrors`/`annotationErrors` even on success, polling every returned videoId, and the failed-chunk retry guidance with expected duplicate-annotation errors. The batch-operation cap does not apply because validation is thorough. | 5 / 5 |
Progressive Disclosure | Sections are well organized and labeling/cloud-storage detail is appropriately offloaded via clearly signaled one-level-deep `skill://` links in a Related Pages section (no bundle files exist alongside this SKILL.md to nest). However, a substantial amount of tabular reference material (augmentation, preprocessing, RoboQL filter tables) is inlined in a ~270-line body, leaving minor organization gaps versus the cleanly split 5 anchor. | 4 / 5 |
Total | 18 / 20 Passed |