Content
31%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 a persona document padded with tool taxonomies and redundant restatements rather than an operational skill: it tells Claude what to be, not concretely what to do. The trigger/scope sections and rough four-step methodology are its strongest parts, but there is no executable guidance, no validation loop, and no progressive disclosure of the ~100-line capability list.
Suggestions
Cut the Capabilities, Knowledge Base, and Behavioral Traits sections (or move a condensed tool reference to a separate references/ file linked from a short overview) — Claude already knows these tools, and the listing consumes most of the token budget.
Replace the generic Instructions with a concrete, sequenced workflow: specific first commands (e.g. an example flamegraph/trace query, a k6 load-test snippet, a Prometheus query for the baseline) and an explicit validate-then-iterate step such as "re-measure after each change and roll back if the budget is exceeded".
Merge the duplicated Purpose/Response Approach/Instructions sections into one workflow and add a concrete worked example (one end-to-end bottleneck diagnosis with commands and before/after metrics) in place of the prompt-only "Example Interactions".
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The Capabilities section spends ~90 lines enumerating tools Claude already knows ("DataDog APM, New Relic, Dynatrace, AppDynamics, Honeycomb, Jaeger"; "k6, JMeter, Gatling, Locust, Artillery"), and Purpose, Knowledge Base, and Behavioral Traits largely restate one another. This matches anchor 2 (noticeably verbose, several padded sections); it is not 1 because it pads with taxonomy rather than explaining known concepts, and the Use/Do-not-use and Instructions sections carry real weight. | 2 / 5 |
Actionability | Guidance like "Collect traces, profiles, and load tests to isolate bottlenecks" and "Propose optimizations with expected impact and tradeoffs" is high-level direction with no commands, tool invocations, code, or worked examples — the "Example Interactions" are only quoted prompts, not demonstrations. This matches anchor 2 (high-level hints missing the specific steps to execute); not 3 because there is no partially executable detail anywhere, and not 1 because a usable rough methodology (baseline → isolate → propose → verify) is present. | 2 / 5 |
Workflow Clarity | The Instructions give a coherent 4-step sequence ("Confirm performance goals... Collect traces... Propose optimizations... Verify results and add guardrails") and the Safety section adds production-guardrail guidance. However, validation is implicit and generic ("verify results") with no error-recovery or re-measurement loop. This matches anchor 3 (sequence present but checkpoints missing or implicit); not 4 because checkpoints lack specificity, not 2 because the ordering is clear and safety steps exist. | 3 / 5 |
Progressive Disclosure | There are no bundle files and no references at all; the ~100-line Capabilities taxonomy is inlined material that clearly belongs in a separate reference file or should be cut. This matches anchor 2 (content that clearly belongs in separate files is inlined); not 3 because there are no references to signal, and not 1 because section headers do give the document navigable structure rather than a monolithic wall. | 2 / 5 |
Total | 9 / 20 Passed |