Content
81%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 delivers an exceptionally clear, gated 8-step design workflow with concrete examples, exact paths, and an executable significance command, backed by a real one-level-deep reference file for the binding schema. Its weakness is token efficiency: scope guardrails and sibling routing are restated four to six times each, and the body carries contract detail that belongs in the reference file. Tightening the repetition would bring this to top-tier without losing clarity.
Suggestions
State the out-of-scope routing once (in the Scope guard) and remove its repeats from Data Sources, step 7, and Reference Materials — the same four sibling handoffs currently appear six times.
Move the artifact-semantics rules from step 8 and the Skill Contract (measurement-contract vs evidence-observation distinctions, 'never call the message validated') into references/stimulus-binding.md, keeping one summary sentence plus the link in the body.
Consolidate the 10-item Reference Materials list with the Next Best Skill section so each sibling skill is linked exactly once instead of appearing in both.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | No concept-teaching padding and every sentence carries information, but the same guardrails are repeated many times — the execution/analysis/authoring handoff routing appears in the intro, Scope guard, Data Sources, step 7, Reference Materials, and Next Best Skill, and "never call the message validated without a matching observation" appears three times. This is 'mostly efficient but could be tightened' rather than the minor-trim level of 4. | 3 / 5 |
Actionability | For an instruction-only skill the guidance is fully actionable: three copy-paste Quick Start invocations, a concrete example hypothesis with a numeric bar ("≥70% of the target panel correctly restate the core benefit unaided after 5 seconds"), a concrete panel-size heuristic labeled Estimated, exact save paths (memory/narrative/message-test-designer/YYYY-MM-DD-<topic>.md), and an executable keyless command ("python3 \"${CLAUDE_PLUGIN_ROOT}/scripts/connectors/experiment.py\" proportion --control <pref_A> <n> --variant <pref_B> <n>"). Specific examples cover the common cases (tagline, hero headline, three positioning statements). | 5 / 5 |
Workflow Clarity | Eight steps are clearly sequenced with explicit validation gates and feedback loops: step 1 stops with NEEDS_INPUT and routes away if no candidate message exists, step 5 scans every claim and tags unapproved ones [needs source], step 6 defines automatic failure routing (a failed test routes back to message-system-architect, not more spend), and the Skill Contract's 'Done when' clause acts as a final checklist. This is a design-and-handoff skill, so no destructive/batch cap applies. | 5 / 5 |
Progressive Disclosure | Good structure: the body is a sectioned overview, and the one real bundle reference (references/stimulus-binding.md — verified to exist) holds the detailed binding schema, well-signaled in step 5, the 'Done when' clause, and a Reference Materials entry with a one-line description. It falls short of 5 because the body is denser than a pure overview (contract semantics and artifact-type rules in step 8 could partly live in the reference file) and ~10 relative links point outside the bundle to sibling skills and repo-level files, which is a minor navigation/organization gap. | 4 / 5 |
Total | 17 / 20 Passed |