Content
85%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-architected skill body with executable guidance, strong destructive-operation checkpoints, and clean one-level-deep progressive disclosure. Its only weakness is inline version/date baseline detail that the rubric expects to live in a centralized or deprecated section rather than the main overview.
Suggestions
Move the version baseline (dxpy==0.410.0, dxCompiler 2.17.0, 2026-07-23 verification date) into references/sources.md and keep only a short pointer in the body, so time-sensitive detail does not bloat the overview.
Consider condensing the 'Current Platform Guidance' bullet list into the relevant reference files (configuration.md, job-execution.md) and retaining only the most load-bearing rules inline.
Centralize deprecated-pattern notes (e.g. top-level `resources` / `runSpec.systemRequirements`) in a single 'Deprecated' subsection of configuration.md rather than stating them inline.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence with no concept-explanation padding, but it carries inline time-sensitive specifics (dxpy==0.410.0, dxCompiler 2.17.0, a 2026-07-23 baseline date) and a broad 'Current Platform Guidance' list rather than centralizing that detail in a deprecated/baseline section, which the rubric says should penalize conciseness. | 2 / 3 |
Actionability | It provides fully executable, copy-paste-ready commands and code — pinned installs, `dx run` with `--input-json-file --destination --cost-limit`, a complete dxpy `find_data_objects` call with real kwargs, and scripted validator invocations with flags. | 3 / 3 |
Workflow Clarity | Risky batch/destructive operations are sequenced with explicit checkpoints: a read-only Safe Preflight, an Operating Contract requiring confirmation before billable/destructive actions, 'show resolved IDs and impact before destructive operations', and feedback loops (re-describe remote state before classifying failure; treat malicious-file warnings as a stop condition). | 3 / 3 |
Progressive Disclosure | SKILL.md is a clear overview with a 'Choose the Right Path' routing table and a Reference Index pointing one level deep to nine real reference files plus two bundled scripts, all verified to exist and clearly signaled, with detail appropriately split out of the main file. | 3 / 3 |
Total | 11 / 12 Passed |