Content
71%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, actionable recon workflow with executable scan commands and clear sequencing. Its main defects are duplicated funnel-stage content across steps, no validation checkpoints for the analysis fields, and — most significantly — critical references (docs/output-kit.md, /atlas-report) that are missing from the bundle, breaking the mandated output-format path.
Suggestions
Bundle docs/output-kit.md (or inline its output-format rules) — the body instructs Claude to follow it, but the file does not exist in the skill bundle, so the required output format is unreachable.
Clarify or provide fallback instructions for the /atlas-report delivery path (e.g., what to do when that skill is unavailable and output exceeds the 40-line CLI budget).
Add concrete detection commands for instrumentation (e.g., grep patterns for analytics/event-tracking calls) so fields like "Tracked?" and "Instrumented" can be answered with evidence rather than guesswork.
Deduplicate the funnel-stage table between Step 1 and the Step 5 output template to reduce token cost.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean — terse scan commands, tables, and bullet checklists with no explanations of concepts Claude already knows. Minor tightening is possible: the funnel-stage table in Step 1 is largely repeated in the Step 5 output template, Step 4's health table overlaps it, and the persona line ("You are Surge — the growth engineer on the Product Team") adds little — matching anchor 4 ("efficient; minor instances that could be trimmed") rather than the every-token-earns-its-place anchor 5. | 4 / 5 |
Actionability | Step 0 provides copy-paste-ready bash find/grep commands, and Steps 1-5 supply concrete fill-in table templates and an output skeleton with explicit placeholders. It falls short of anchor 5 because the analysis steps (1-4) give no concrete detection method for key fields like "Tracked?" / "Instrumented" (e.g., no commands for finding analytics/event instrumentation), leaving minor execution gaps. | 4 / 5 |
Workflow Clarity | A clear, well-sequenced six-step process (Step 0 detect environment → Step 5 present assessment) with a defined output contract. Operations are read-only reconnaissance, so the destructive/batch validation cap does not apply, but there are still no explicit validation checkpoints or feedback loops (e.g., verifying a stage's "Tracked?" answer before filling the assessment), which keeps it at anchor 4 rather than 5. | 4 / 5 |
Progressive Disclosure | Sections are well organized and the skill is largely self-contained, but the body depends on two references that do not exist in the bundle: "Follow the output format defined in docs/output-kit.md" (no docs/ directory exists) and "invoke /atlas-report" (not part of this skill). References that are broken rather than clearly signaled, plus a >50-line body that inlines the full output template, match anchor 3 ("references present but not clearly signaled; content that should be separate is inline"). | 3 / 5 |
Total | 15 / 20 Passed |