Content
75%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 well-structured, information-dense reference that consistently provides concrete tool names, commands, and examples, with genuine error-recovery guidance in the upload pipelines. It uniformly falls just short of top marks: a few workflows lack verification checkpoints, several prose passages could be tighter, and the large reference tables sit inline rather than in separate files.
Suggestions
Add an explicit verification step to the version-creation workflow (e.g., confirm generation succeeded via versions_get and check the version's image counts before announcing completion), matching the validate-then-proceed pattern already used in the upload flows.
Trim hedge-heavy conditional prose in the upload sections (e.g., 'An authentication, validation, or server error alone does not establish that the bridge is unavailable') to terse decision rules, preserving the trigger conditions but cutting explanation.
Show a minimal call shape for the key MCP tools referenced but never exemplified (annotations_save with add_to_dataset=true, versions_generate with preprocessing options) so agents can execute them without guessing arguments.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Dense, table-driven content where nearly every row carries non-obvious platform facts (limits, tier gating, split-folder aliases), and prose encodes real gotchas (dedup IDs, chunk-retry semantics). A few hedge-heavy sentences — 'An authentication, validation, or server error alone does not establish that the bridge is unavailable' and the long annotation-group trap paragraph — could be tightened. Not 5: some sentences over-explain conditional reasoning. | 4 / 5 |
Actionability | Concrete executable artifacts throughout: 'roboflow import -w <workspace> -p <project-id> /path/to/dataset', 'project.version(1).download("yolov8")', exact MCP tool names, and specific filter examples like 'filename:*factory*' and 'class:helmet>=3'. Not 5: tools like 'annotations_save' and 'versions_generate' are named without call shapes or argument examples, leaving minor gaps. | 4 / 5 |
Workflow Clarity | The upload workflows have explicit validation and recovery loops — poll 'image_upload_status' until 'completed' or 'failed', inspect 'result.uploadErrors', re-poll every 'result.videoIds' entry, and the failed-chunk retry guidance — satisfying the batch-operation validation requirement. Not 5: the 5-step version creation pipeline and the export flow lack verification checkpoints (e.g., confirming via versions_get after generate). | 4 / 5 |
Progressive Disclosure | Well-organized sections with clearly signaled one-level references (Related Pages pointing to the labeling and cloud-storage skills), appropriately splitting labeling and cloud-storage detail out of this file. Not 5: at ~245 lines, the full preprocessing and augmentation tables are inlined in SKILL.md where a reference file could carry the bulk; not 3: structure and signaling are good. | 4 / 5 |
Total | 16 / 20 Passed |