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.
The body is an efficient, well-sequenced design playbook with genuinely non-obvious domain notes and good error-recovery guidance, but it stops at design: no tool call, command, or config example shows how to actually apply any optimization, and there is no step to verify improvements after changes are made. Single-file organization is solid with minor room to split per-topic detail into references.
Suggestions
Add concrete apply-path guidance: for each optimization step, show the MCP tool call (e.g., a harness pipeline-update call with resource_type and YAML payload parameters) or an example Harness pipeline YAML snippet for enabling Test Intelligence, parallelism, and a cache step — the read-only call blocks already model the format.
Close the workflow with a validation checkpoint: after applying changes, re-run the pipeline, pull execution timings with the harness_get call from Step 5, and compare against the Step 1 baseline to confirm the time savings before reporting done.
Consider moving per-build-tool caching details and the monorepo dependency-graph guidance into one-level-deep reference files so SKILL.md stays a shorter overview that branches by optimization type.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean, imperative, bullet-driven content with zero padding or explanation of concepts Claude already knows; the explanatory asides that exist are genuinely non-obvious domain knowledge ('Test Intelligence needs 2-3 full test runs to build its initial model', 'Cache keys should be based on lockfiles, not timestamps', 'Set cache TTL (recommended 7-14 days)'). Every section adds information an agent would not already have. The Examples section slightly restates step content, but each line pairs a user phrasing with the matching response path, which earns its place. | 5 / 5 |
Actionability | Data-gathering steps are executable — two harness_list call blocks and a harness_get block with concrete parameters — but the configuration core of the skill (Steps 3, 4, 6) provides only design parameters ('Enable Test Intelligence in the pipeline stage', 'Configure test splitting method: by class, by file, or by timing data', 'Use path-based triggers: only build services with changed files') with no tool call, command, or YAML example for actually applying any of it. This matches 'some concrete guidance but incomplete; missing key details' better than anchor 4, where gaps would be minor — here the apply-the-change path is entirely absent. Anchor 5 is clearly out of reach: only read-only calls are templated, nothing for configuration. | 3 / 5 |
Workflow Clarity | Clear six-step sequence: scope and baseline timing data first (Step 1), task identification branching into five paths (Step 2), then per-task design steps, with a Troubleshooting section providing error-recovery loops for failure modes (TI not skipping tests, cache misses, monorepo building everything). Not 5: no post-change validation step — nothing instructs re-running the pipeline and comparing stage timings to confirm the optimization actually helped, which is the natural closing checkpoint for an optimization skill. Not 3: checkpoints are mostly present (baseline gathering up front, troubleshooting recovery at the back) and the gap is a single missing verify step. | 4 / 5 |
Progressive Disclosure | Well-organized single-file skill with clear step, example, performance-note, and troubleshooting sections; the roughly 150-line body stays navigable with no nested or buried references. Not 5: the five optimization types are each self-contained playbooks that could live in one-level-deep reference files to keep SKILL.md as a shorter overview (e.g., per-build-tool caching details), which is a minor organization gap rather than 'content appropriately split'. | 4 / 5 |
Total | 16 / 20 Passed |