Content
66%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 well-structured and highly actionable with a clear sequenced workflow, explicit validation/NEEDS_INPUT checkpoints, and correct use of a single one-level-deep reference file. Its main weakness is token efficiency: the identity-ref and authorization contract is repeated roughly four times across the Skill Contract and Instructions, consuming a large share of the body without adding new guidance.
Suggestions
State the opaque-ref and exact-authorization rules once in a single canonical section (e.g., the Writes bullet) and cross-reference it from Done-when #5 and Instructions 1 and 6 instead of restating it verbatim — this alone would cut roughly a third of the body.
Trim step 6 to the essential decision points (offer WARM save, transient render rule, distribution needs fresh authorization) and drop the repeated enumeration of every ref kind, which already appears in the Skill Contract.
Compress the Reference Materials and Next Best Skill lists into one compact handoff line per skill; eight-plus deep relative links (../../../...) are unverifiable from the bundle and dilute the single genuine bundle reference.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Noticeably padded: the opaque-ref/authorization rules are restated at least four times ("Every saved report, template field, appendix, and handoff uses opaque `client_ref`..." in Writes; again in Done-when #5; "Keep any raw identity/contact locator transient" in step 1; and the full restatement in step 6), and the Promotes bullet plus the paid-ads blockquote add contract boilerplate that crowds out report-writing guidance. It is not a 3 because the duplication is substantial rather than a few stray sentences that could be tightened. | 2 / 5 |
Actionability | Concrete, executable guidance for an instruction-only skill: Quick Start invocations, seven specific steps with fallback behavior ("If the required computed artifact is absent, return a partial shell plus `NEEDS_INPUT` and the exact handoff; do not recompute"), and a worked Example with a filled metrics table. Not a 5 because the in-body example covers only the executive audience; client/team cases live entirely in the reference file. | 4 / 5 |
Workflow Clarity | Seven clearly numbered steps (determine parameters → pick template → apply guidance → close with recommendations → preserve Retro Card → return/persist → offer next skill) with explicit validation branches for every failure mode: absent artifacts yield a "partial shell plus `NEEDS_INPUT`", and an "absent, mismatched, or stale" Retro Card yields a "bound `decision: unknown` placeholder". The five-item "Done when" checklist closes the loop with verifiable exit criteria. | 5 / 5 |
Progressive Disclosure | Good structure: 543 lines of full templates are correctly split into a real, one-level-deep, well-signaled reference ("full executive, client, and internal-team templates live in [report-templates.md](references/report-templates.md)") cited consistently. Not a 5 because the body carries 8+ out-of-bundle relative links with deep paths (e.g., ../../../seo-geo/tune/content-quality-auditor/SKILL.md) that cannot be verified from the bundle and add navigation clutter. | 4 / 5 |
Total | 15 / 20 Passed |