Content
75%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, highly actionable tool skill: it sequences install through verdict across nine steps with concrete commands and near-complete YAML for the main chaos types, includes verification and a useful anti-patterns table. The main gaps are marketing-quote padding, placeholder selectors in the Workflow example, a referenced verdict script that does not exist in the bundle, and no error-recovery guidance.
Suggestions
Remove the quoted homepage passages ("enormous capability to orchestrate fault scenarios", "no special dependencies required", etc.) and state the facts directly — Claude does not need the marketing framing, and this would lift conciseness to 5.
Complete the Workflow example by replacing the "selector: { ... }" placeholders with real selectors, and either inline the steady-state verdict check in the CI step or ship the referenced ./scripts/datadog-verdict.sh, since the file does not exist in this bundle.
Add an error-recovery checkpoint after Step 7 (e.g., what to do when an experiment never transitions from Created to Running, and how to clean up on abort) to close the workflow feedback-loop gap for these destructive operations.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly lean (commands, YAML manifests, compact tables), but quoted homepage passages like "enormous capability to orchestrate fault scenarios" and "no special dependencies required" are marketing filler, and the pinned "--version 2.6.3" is time-sensitive detail not placed in a deprecated/old-patterns section — efficient with minor trimming possible, so 4 rather than 5. | 4 / 5 |
Actionability | Guidance is mostly executable and copy-paste ready (helm install with pinned version, kubectl port-forward/apply/get/describe, three near-complete YAML manifests), but the Workflow example uses "selector: { ... }" pseudocode placeholders and the CI step references "./scripts/datadog-verdict.sh", which does not exist in this bundle — minor gaps, so 4 rather than 5. | 4 / 5 |
Workflow Clarity | Steps 1-9 give a clear install → author → orchestrate → run/verdict sequence with explicit verification ("kubectl get networkchaos -w", "verify the steady-state hypothesis held", CI Datadog check), so the destructive-operation cap does not apply; however there is no error-recovery feedback loop (e.g., what to do if an experiment stays in Created, or cleanup on abort), matching 4 rather than 5. | 4 / 5 |
Progressive Disclosure | The 255-line body is well-sectioned with clear headers and a References section, but all detail is inline — no local bundle files exist, and the only references are external URLs plus pointers into a sibling skill's files; content like the full Workflow and PhysicalMachineChaos manifests could be offloaded to reference files. Good structure with minor organization gaps, so 4 rather than 5 (the under-50-lines exception does not apply). | 4 / 5 |
Total | 16 / 20 Passed |