Content
68%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-structured, lean operating manual with concrete tool references and clear per-mode sequences, but its destructive reconciliation workflow relies on implicit rather than explicit validation checkpoints.
Suggestions
Add an explicit verification step after reconciling @agent elements (e.g., 'Re-read each cited element and confirm no @agent text remains before finishing').
Consolidate the repeated tool inventory and 'read the board first' guidance into the shared-board section, referencing it from each mode to remove repetition.
Include one short example of a watch-mode response and a NOTHING_TO_ADD turn to make the actionable guidance copy-paste concrete.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence, never explaining Excalidraw basics; the only trimming opportunity is mild repetition of the tool inventory and the 'read the board first' instruction across modes. | 4 / 5 |
Actionability | Provides concrete tool names (create_element, update_element, describe_scene, etc.), an exact sentinel ('NOTHING_TO_ADD'), and explicit reconcile rules; minor gaps are the absence of example turn outputs and assumed MCP tool signatures. | 4 / 5 |
Workflow Clarity | Each mode has a clear sequence with an implicit verification ('no @agent text remains... or it will re-trigger'), but the destructive element reconciliation lacks an explicit validation checkpoint, so the destructive-operation cap of 3 applies. | 3 / 5 |
Progressive Disclosure | Content is self-contained with a clean section hierarchy (shared board, two modes with subsections, etiquette) and needs no external files; it sits slightly over the ~50-line simple-skill threshold, keeping it just below a 5. | 4 / 5 |
Total | 15 / 20 Passed |