Content
73%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 strong, highly structured instruction skill: an unambiguous eight-step workflow with NEEDS_INPUT error handling, dry-run-by-default mutation, a done-when checklist, and concrete segment definitions and commands. The main costs are cross-section repetition of the consent-registry and sibling-skill disclaimers, and heavy dependence on external files for output templates that are not part of this bundle.
Suggestions
State the 'reads memory/consent/, never writes it' rule once (e.g. in Data Sources or Scope guard) and reference it elsewhere in a few words instead of restating the full caveat four times.
Collapse the overlap between Reference Materials, Scope guard, and Next Best Skill — each currently re-lists email-creative-builder, email-sequence-designer, and consent-registry with near-identical annotations; one routing section would tighten the body noticeably.
Inline a minimal example of the segment-map / handoff-summary output shape (or include it as a bundled reference file) so the expected artifact does not depend solely on the external ../../../references/skill-contract.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious domain detail, but the consent-registry caveat "this skill reads it, never writes it" / "never writes `memory/consent/`" is restated at least four times (Data Sources, Skill Contract Promotes, Instructions step 6, Reference Materials), and the sibling-skill disclaimers are repeated across the intro, Scope guard, Reference Materials, and Next Best Skill — matching "mostly efficient but... could be tightened". Not a 4: the redundancy is pervasive across sections rather than a few minor trimmable instances; not a 2 because nothing explains concepts Claude already knows and every section carries real information. | 3 / 5 |
Actionability | Concrete, executable guidance throughout: named segment definitions tied to exported columns ("engaged-90d = opened or clicked in last 90 days", "cart-abandoners-7d"), copy-paste commands ("python3 \"${CLAUDE_PLUGIN_ROOT}/scripts/connectors/resend.py\" contacts"), an explicit save path template, and freeze fields (segment_ref, definition_version, definition_hash, evaluated_at). Not a 5: the actual output artifact shapes (handoff summary format, save-results template) are delegated to external skill-contract.md rather than shown, so a reader cannot fully produce the expected output from this file alone; not a 3 because what is present is executable, not pseudocode or vague hints. | 4 / 5 |
Workflow Clarity | Eight clearly sequenced steps with explicit validation checkpoints and error handling: missing columns "become NEEDS_INPUT flags, not guesses", step 6 reconciles the suppression list against consent/suppression snapshot refs, step 7 freezes definitions with version/hash and evaluated_at, mutating connector subcommands are "dry-run by default (--live to execute)", and a "Done when" checklist closes the loop (every size labeled, no raw PII in artifacts, reconciliation or NEEDS_INPUT). This matches the anchor with explicit validation steps, error-recovery paths, and a checklist; a 4 would require missing or merely implicit checkpoints, which is not the case. | 5 / 5 |
Progressive Disclosure | The body is a well-sectioned ~75-line overview keeping the core workflow inline and pointing outward with clearly annotated, one-level-deep links (send-benchmark.md, CONNECTORS.md, SECURITY.md, send-control.md, sibling SKILL.md files). Not a 5: no bundle files exist here (references/, scripts/, assets/ are absent), so the many ../../../ references — including the output templates the skill depends on — cannot be verified and pull essential detail outside this skill's own bundle; not a 3 because the structure and signaling are genuinely good, with nothing inlined that clearly belongs in a separate file. | 4 / 5 |
Total | 16 / 20 Passed |