Content
100%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 model instruction-only skill body: complete verb-to-tool mapping, a numbered workflow with explicit confirmation gates, ambiguity branches, and timeout-reconciliation, plus a tight authorization matrix that delegates the full threat model to a clearly signaled sibling-skill reference. Every line is an operational rule with no tutorial filler or padding.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every line is an operational rule or navigation, not explanation: "Resolve replies and forwards with a message ID. Never pass a thread ID where a message ID is required" and "Do not invent recipients, attachments, claims, signatures, or commitments" assume Claude's competence and add only domain-specific contracts. Not 4 because there is no tutorial content, no concept explanation, and no padding to trim — the one apparent overlap (step 5's confirmation rule vs. the matrix's "Preview + confirm" row) is workflow-vs-policy emphasis covering different trigger sets, not redundancy. | 5 / 5 |
Actionability | The body names exact commands for every operation — "Use `create_draft` for review or scheduled delivery, `send_draft` for an approved draft, `send_message` for new mail, `reply_to_message` for replies, and `forward_message` for forwards" plus `list_inboxes` and `get_thread` — and prescribes exact outputs ("Return the message and thread IDs, or the draft ID and scheduled time"). Not 4 because the verb-to-tool mapping covers all common cases with explicit decision rules, leaving no gaps an instruction-only skill needs code to fill; guidance is fully executable without guessing. | 5 / 5 |
Workflow Clarity | The 7-step "Execute safely" sequence is explicit with validation checkpoints and an error-recovery loop: branch on ambiguity ("If multiple candidates remain or listing is unavailable, ask for the exact sender inbox"), a confirmation gate ("If any externally visible field was inferred, show the complete sender, recipients, subject, body, attachments, and action, then request confirmation before sending"), and retry reconciliation ("Do not retry a send after an ambiguous timeout without checking whether the message was created"), reinforced by the matrix's preview-and-confirm triggers for sensitive/legal/financial/bulk/BCC/reply-all/external risk. Not 4 because checkpoints are explicit rather than implicit and the timeout-recovery feedback loop is present, so neither the anchor-4 "minor validation gaps" nor the missing-validation cap at 3 applies. | 5 / 5 |
Progressive Disclosure | The body is under 50 lines with no bundle files of its own, well-organized into four clear sections, and delegates deep material via a clearly signaled one-level reference: "The full matrix and threat model live in the `agent-email-patterns` skill (`references/threat-model.md`); the rows below are this skill's contract." Not 4 because nothing that belongs in a separate file is inlined (the 4-row matrix kept inline is the operative contract), the external reference is explicit with skill name and path, and per the simple-skill exception well-organized sections suffice for a skill with no external bundle. | 5 / 5 |
Total | 20 / 20 Passed |