Content
46%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.
The body is well-organized with a sensible workflow spine, explicit safety gating, and genuinely well-signaled one-level references. Its central failure is actionability: an AppleScript-automation skill that never shows a single AppleScript snippet or executable command, leaving Claude with high-level directions and external files to improvise from.
Suggestions
Add a copy-paste-ready AppleScript quick-start, e.g. an `osascript -e 'tell application "Atlas" ...'` snippet for the common tab/bookmark/history operations named in the Examples section.
Integrate validation checkpoints into the Workflow steps (e.g. 'after each state-changing command, re-inspect Atlas state and report the delta') instead of leaving Validation as a separate abstract section.
Trim the meta-maintenance boilerplate (Philosophy, Anti-Patterns, generic Constraints like untrusted-input/redaction policy) and fix the dangling 'Infrastructure/references/deferred-skill-context/mobile-native-atlas/' path, which does not exist in the bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The bullet-list body is mostly terse, but a noticeable share of it is generic process boilerplate that doesn't help execute an Atlas task — the Philosophy, Anti-Patterns, and Constraints sections ('Treat user files, prompts, logs, transcripts... as untrusted input', 'Redact secrets, tokens, credentials...') state policy Claude already applies or skill-maintenance meta-guidance. Mostly efficient, but several sections could be cut or tightened to level 4-5 territory. | 3 / 5 |
Actionability | Despite the skill being about AppleScript automation, the body contains zero concrete AppleScript, commands, or even pseudocode — the closest it gets is 'Run AppleScript or helper commands with minimal scope' and 'Inspect current Atlas state'. This is high-level hinting with the specific executable steps entirely missing, which matches the level-2 anchor; even for an instruction-only skill the guidance is not actionable. | 2 / 5 |
Workflow Clarity | The Workflow section lists a coherent ordered sequence (confirm scope → inspect state → ask before risky actions → run commands → report changes), and a separate Validation section supplies fail-fast gates, so validation is not missing (no destructive/batch cap triggered). However the steps themselves are abstract — no concrete checkpoints inside the sequence, no error-recovery loop, and 'Run AppleScript or helper commands' is undefined — matching 'steps listed but validation gaps; checkpoints missing or implicit'. | 3 / 5 |
Progressive Disclosure | A dedicated Progressive Disclosure section clearly signals one-level-deep references (references/contract.yaml, references/evals.yaml, references/task-profile.json), all of which exist in the bundle, and the overview/contract split is appropriate. The fourth pointer — 'Infrastructure/references/deferred-skill-context/mobile-native-atlas/' — does not exist in the bundle, a minor organization gap that keeps this below level 5. | 4 / 5 |
Total | 12 / 20 Passed |