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.
An exemplary hub-style SKILL.md: lean, entirely tool-specific, with executable discovery commands, a clear starting point, and condition-based routing to real reference files. The only structural note is the second level of nesting under build-workflow.md, which keeps progressive disclosure just below the top anchor.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every sentence carries tool-specific knowledge Claude cannot infer (dotted operation ids, --json/DIFY_OUTPUT=json forcing, --all/--full semantics, pins, stderr-on-logout behavior). There is no padding and no explanation of concepts Claude already knows. | 5 / 5 |
Actionability | The body gives copy-paste-ready commands for every discovery path — `difyctl help --json`, `difyctl help <namespace> --json`, `difyctl help upload file`, `<command> --help` — plus concrete output expectations (top 20 entries plus total, input/bind/kind/examples/pins). | 5 / 5 |
Workflow Clarity | A clear, ordered discovery sequence with an explicit 'Start here' entry point, error-recovery guidance ('Widen the words when nothing fits', not-logged-in/stderr behavior), and an explicit safety checkpoint for destructive operations ('Operations that remove or revoke have no confirmation prompt. Confirm with the human first'). | 5 / 5 |
Progressive Disclosure | The 'Where to read next' section links to four real, well-signaled top-level references (setup.md, operations.md, build-workflow.md, plugins.md) each with a clear trigger condition, and the body itself stays a lean overview. However, the bundle nests a second level (references/build-workflow.md links to six files under references/build-workflow/), so references are not strictly one level deep. | 4 / 5 |
Total | 19 / 20 Passed |