Content
77%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This is a well-structured, highly actionable skill that clearly defines a diagnostic-first workflow for prompt improvement. Its main strength is the concrete process with explicit gates preventing premature changes, and the issue-to-fix mapping table. Its main weakness is moderate verbosity — the 'Core Problem' section and the inlined best practices in Step 2 add length that could be trimmed or offloaded to the referenced file.
Suggestions
Move the detailed best practices content in Step 2 entirely to the referenced `references/anthropic-best-practices.md` file, keeping only a brief summary or the most critical 2-3 points inline to reduce token usage.
Consider condensing or removing the 'Core Problem' section — Claude doesn't need to be told what agents do wrong in general; the 'Stop If You Catch Yourself' table already covers this more actionably.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The skill is mostly efficient but includes some unnecessary content. The 'Core Problem' section explains failure modes Claude should already understand, and the 'Stop If You Catch Yourself' table, while useful, is verbose. The quick reference table at the end is lean and valuable, but the middle sections could be tightened. | 2 / 3 |
Actionability | The skill provides highly concrete, actionable guidance: a clear required process with specific steps, a diagnostic table mapping issues to fixes, explicit examples of what counts vs. doesn't count as a concrete issue, and before/after framing for proposed changes. The instructions are specific enough to be directly followed. | 3 / 3 |
Workflow Clarity | The four-step process is clearly sequenced with explicit gates (Step 1 requires concrete issues before proceeding, Step 4 requires rationale for each change). The 'Stop If You Catch Yourself' table serves as a validation checkpoint. The workflow has a clear feedback loop: if no concrete issue is identified, stop and ask. | 3 / 3 |
Progressive Disclosure | The skill references `references/anthropic-best-practices.md` for the full reference, which is good progressive disclosure. However, no bundle files are provided, so this reference is unverifiable. Additionally, Step 2 inlines a substantial amount of the best practices content that could arguably live entirely in the referenced file, making the main skill longer than necessary. | 2 / 3 |
Total | 10 / 12 Passed |