Content
78%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 compact, well-structured protocol skill: it gives concrete trigger conditions, a fully enumerated type parameter with examples, and explicit edge-case rules without any padding or generic explanations. The only meaningful gaps are the absence of a sample working_memory_note invocation and a slightly over-detailed MEMORY.md taxonomy that could be trimmed to the bullets actually used.
Suggestions
Add a short example working_memory_note call (with a type argument) so the exact invocation format is copy-paste ready.
Trim the MEMORY.md taxonomy bullets to those the protocol actually uses (notably Crystal Pointers and Active Context) to save tokens.
State a one-line tie-break rule for when multiple triggers or types apply (e.g., prefer directive over experience for corrections).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean with no padding and no explanations of concepts Claude already knows — the dream engine and MEMORY.md structure are bespoke system context, not general knowledge. The 7-bullet MEMORY.md taxonomy is partially dispensable (only 'Crystal Pointers' is referenced later), which keeps it below anchor 5, but each bullet is a compact one-liner earning partial relevance, matching anchor 4's 'minor instances of over-explanation'. | 4 / 5 |
Actionability | Concrete, executable guidance throughout: specific trigger conditions, the full `type` parameter enumeration with inline examples ('I prefer X', 'always do Y', 'never Z'), the Crystal Pointer line syntax (`-> search: \`keywords\``), and the target file `memory/scratch.md`. It stops short of anchor 5 only because it lacks an example `working_memory_note` call showing the exact invocation format. | 4 / 5 |
Workflow Clarity | A simple decision-protocol skill with unambiguous if-then rules: triggers for noting, type selection with a stated default (`experience`), and explicit edge-case handling for Crystal Pointer recall and profile corrections. No destructive or batch operations, so no validation cap applies; it misses anchor 5 only because no explicit ordering is given for cases where multiple triggers or types could apply. | 4 / 5 |
Progressive Disclosure | Under 50 lines with no need for external reference files, and the content is cleanly sectioned with descriptive headers ('When to use working_memory_note', 'Epistemic type parameter', 'Crystal Pointers', 'Correcting the user profile'). Per the rubric's simple-skill note, well-organized sections alone warrant a 5 here; there are no bundle files, so there is nothing to over- or under-disclose. | 5 / 5 |
Total | 17 / 20 Passed |