Content
62%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body provides a clear, well-sequenced workflow with strong approval and revision checkpoints, and gives specific MCP tool guidance. Its weaknesses are mild redundancy across sections, prose-described rather than fully templated drafting examples, and a monolithic single-file structure with no external references.
Suggestions
Consolidate the inbox-filtering rule and voice/tone guidance into a single location and cross-reference it, removing the near-duplicate explanations between Step 2/Step 4 and the Important guidelines section.
Add concrete, fill-in-the-ready example `instructions` strings (beyond the one Jake example) so drafts are copy-paste actionable rather than described in prose.
Move the per-tool parameter reference and detailed voice/tone matching notes into a separate reference file and link to it from the body, achieving one-level-deep progressive disclosure.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient and actionable but repeats guidance — the `labels: ["INBOX"]` rule appears in Step 2 and again nearly verbatim in Important guidelines, and voice/tone matching is explained in both Step 4 and the guidelines — so it could be tightened rather than hitting the lean level-3 anchor. | 2 / 3 |
Actionability | It names concrete MCP tool calls with exact parameters (e.g., `Superhuman_Mail.list_threads` with `is_unread: true` and `labels: ["INBOX"]`, `create_or_update_draft` with `instructions`), but the drafting guidance is prose-described rather than provided as complete copy-paste-ready templates, leaving some key construction detail implicit. | 2 / 3 |
Workflow Clarity | A clear 6-step sequence includes explicit validation checkpoints — the Step 3 approval gate ("Shall I go ahead and draft all of these?"), "Never send without approval," and the Step 6 revision feedback loop — satisfying the batch-operation feedback requirement of the level-3 anchor. | 3 / 3 |
Progressive Disclosure | The skill is well-organized into clear sections but is a single ~109-line file with no bundle files or external references, so content that could be split (e.g., an MCP tool reference or voice-guidance doc) is inline rather than following the one-level-deep reference pattern of level 3. | 2 / 3 |
Total | 9 / 12 Passed |