Content
67%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 a well-sectioned overview with a genuinely executable workflow, a concrete keyless probe command, honest measurement-latency guidance, and a clean one-level-deep local reference bundle. Its weaknesses are token weight spent on cross-suite contract/termination machinery and a wide fan-out of parent-repo links that reduce standalone navigability.
Suggestions
Compress or relocate the "Skill Contract" reads/writes/promotes and termination-note sections (e.g., into the already-referenced skill-contract.md) so SKILL.md spends its budget on GEO guidance.
Inline the Handoff Summary format (or a 3-line skeleton) instead of routing through ../../../references/skill-contract.md, and prune the parent-repo link fan-out (platform refs, llms-txt/OKF) to those the optimizer actually needs.
Add one short fix-and-recheck instruction after Step 5 (e.g., "if any self-check item is Fail, revise that passage and re-score") to close the validation loop explicitly, and inline the one-line meanings of the four gate CORE-EEAT IDs (C02, O03, O05, E01).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The operational core (Quick Start, five-step Instructions, labeling rule) is lean and assumes competence, but a substantial share of tokens goes to host-framework ceremony — the "Skill Contract" reads/writes/promotes block, the "Handoff Summary" pointer, "Promotes ... to memory/hot-cache.md and memory/open-loops.md", and the visited-set/"max handoff depth of 3" termination note — none of which advances the GEO task itself. This fits anchor 3 ("mostly efficient but includes some unnecessary explanation or could be tightened"); it is not anchor 4 because the contract/termination boilerplate is more than a minor trim. | 3 / 5 |
Actionability | Concrete, executable guidance dominates: a copy-paste command ("python3 \"${CLAUDE_PLUGIN_ROOT}/scripts/connectors/tavily.py\" search \"<target query>\" --answer --limit 10"), specific techniques ("standalone 25-50 word definitions, sourced quotable statements ... FAQ schema"), a save-path template ("memory/content/YYYY-MM-DD-<topic>.md"), and the Measured/User-provided/Estimated/N-A labeling rule. It misses anchor 5 because the Example section defers the actual output to a reference file and the CORE-EEAT IDs (C02, O03, E01) are unresolvable without loading the reference. | 4 / 5 |
Workflow Clarity | The five numbered steps form a clear sequence ending in an explicit validation step ("CORE-EEAT GEO Self-Check — verify C02, C04, C09 ... with Pass/Warn/Fail") with a done-when gate ("the CORE-EEAT GEO self-check ... has no unaddressed Fail"), and the operation is not destructive/batch so no cap applies. It sits at anchor 4 rather than 5 because there is no explicit fix-and-re-check loop spelled out in the body (the iterate-until-pass behavior is only implied by "no unaddressed Fail") and the self-check criteria live in the reference. | 4 / 5 |
Progressive Disclosure | Scoring against the actual bundle: all six cited local references exist with real content, are exactly one level deep, and are clearly signaled with per-file one-line descriptions in the "Reference Materials" list — close to anchor 5. It drops to anchor 4 ("minor organization gaps") because the body also fans out to ~10 paths outside the bundle ("../../../references/skill-contract.md", "../../../CONNECTORS.md", "../../../references/platforms/x.md", "../../tune/content-quality-auditor/SKILL.md", "scripts/connectors/*") which are unresolvable within this skill's directory and make the load-bearing Handoff Summary format depend on an external file. | 4 / 5 |
Total | 15 / 20 Passed |