Content
92%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 lean, well-sequenced operational contract: a concrete tool mapping, a numbered workflow with explicit confirmation and retry-reconciliation checkpoints, and sharp authorization/security boundaries. The only weakness is actionability detail — no example tool invocation and an unnamed search tool for post-timeout reconciliation.
Suggestions
Name the specific MCP tool used to reconcile an ambiguous send (e.g., the search tool) instead of the vague 'message/thread/search evidence' phrasing in the authorization matrix.
Add one minimal example invocation (e.g., send_message with sender, recipient, subject, body fields) so common-case guidance is copy-paste ready.
Specify what to do when get_thread does not contain the referenced message ID (e.g., ask the user for the exact message rather than falling back to a thread ID).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | A lean ~42-line body where every section prescribes non-obvious policy — tool-to-operation mapping, message-ID vs thread-ID, inbox resolution, retry reconciliation, untrusted-content rules — with zero space spent explaining concepts Claude already knows. | 5 / 5 |
Actionability | Concrete MCP tool names cover every common case (create_draft, send_draft, send_message, reply_to_message, forward_message, list_inboxes, get_thread) and each ambiguity maps to a specific safeguard, but there are minor gaps: no example invocation or parameter shape, and 'Reconcile via message/thread/search evidence' never names the search tool. Mostly-executable with minor gaps fits the 4 anchor; it falls short of 5's copy-paste-ready coverage. | 4 / 5 |
Workflow Clarity | 'Execute safely' gives a numbered 7-step sequence with explicit validation gates — step 5's full preview-and-confirm when any visible field is inferred, and step 6's no-blind-retry rule — plus a feedback loop for the ambiguous-timeout error case. Checkpoints are explicit, not implied, matching the 5 anchor. | 5 / 5 |
Progressive Disclosure | The skill is under 50 lines, needs no bundle files, and is organized into clear sections (operation choice, execution, authorization, security). Its one external pointer (the agent-email-patterns skill's references/threat-model.md) is clearly signaled and one level deep. | 5 / 5 |
Total | 19 / 20 Passed |