Content
85%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 instruction body with a clear gated workflow and correct use of one-level-deep reference files. The main weaknesses are repetitive restatement of the decision-ownership rule and incomplete spell-outs of two of the three helper subcommands.
Suggestions
State the decision-ownership rule fully once (step 7) and reduce the restatements in the intro, Skill Contract, and Next Best Skill sections to a single short cross-reference, e.g. "per the decision gate in step 7".
Show full invocation examples for experiment.py's continuous and samplesize subcommands alongside the existing proportion example so all three compute paths are copy-paste ready.
Consolidate the duplicate sibling-skill boundary statements (intro paragraph and Next Best Skill) into the Next Best Skill section only.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with no explanations of concepts Claude already knows (formulas and the sizing table are pushed to the reference file), but the decision-ownership rule is restated roughly four times (intro, step 7, step 9, Next Best Skill) and the sibling-skill exclusions appear twice. This matches "efficient; minor instances of over-explanation that could be trimmed" rather than the fully lean 5 anchor. | 4 / 5 |
Actionability | Concrete executable guidance is present: a copy-paste-ready command ("python3 \"${CLAUDE_PLUGIN_ROOT}/scripts/connectors/experiment.py\" proportion --control <conv> <n> --variant <conv> <n> --alpha <alpha> --min-lift <relative-bar>"), a fill-in-the-blank hypothesis template, and explicit defaults ("alpha=.05", "power=.80"). It falls short of 5 because the continuous and samplesize subcommands are named but their invocations are not spelled out like proportion's. | 4 / 5 |
Workflow Clarity | Instructions 1–9 form a clear sequence (mode → hypothesis → matrix → metrics → sizing → significance → decision ownership → provenance → binding verification) with explicit validation gates and feedback paths: the step-1 NEEDS_INPUT stop, step 9's "Refuse to combine a result with a different creative/landing hash... A changed binding starts a new test", and the decision gate in step 7. This matches the anchor for clear sequencing with explicit validation and error-recovery loops. | 5 / 5 |
Progressive Disclosure | The ~78-line body is an overview with well-signaled one-level-deep references: both bundle files exist (references/test-design-guide.md, references/measurement-control.md) and each citation states what it contains ("variant matrix, reference sizing table, statistical procedures, and decision-ownership matrix"). The heavy detail is appropriately split out rather than inlined. | 5 / 5 |
Total | 18 / 20 Passed |