Content
88%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.
This is a well-engineered operational skill: concrete tool guidance, copy-paste repair commands, and unusually thorough error-recovery loops for empty results, server failures, and a corrupt vector index. The main improvement levers are trimming the persona section and the duplicated fact-supersession guidance, and splitting the recovery/coordination detail into reference files now that the body has grown past overview length.
Suggestions
Remove or compress the "Identity" persona section ("Act as a senior AI-memory systems engineer...") — it adds tokens without changing agent behavior, improving the conciseness score.
State the kg_supersede / kg_invalidate / kg_add distinction once (either in Protocol step 6 or in the Conflicting-facts unhappy path, not both) to remove duplication.
Move the corrupt-index recovery procedure and the coordination/logstream guidance into reference files linked from the body, slimming SKILL.md toward an overview with clearly signaled one-level-deep pointers.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with operational specifics (tool table, argument specs, repair commands, unhappy paths), but the "Identity" persona section ("Act as a senior AI-memory systems engineer with decades of experience") is padding, and the kg_supersede/kg_invalidate/kg_add guidance is stated twice (Protocol step 6 and the Conflicting-facts unhappy path) — trimmable, so anchor 4 rather than 5. | 4 / 5 |
Actionability | Fully executable guidance throughout: named MCP tools with argument semantics ("`mempalace_search` takes a short natural-language `query` ... plus optional `wing` / `room` filters and `limit` (default 5)"), copy-paste shell commands (`mempalace --version`, the `mempalace repair --mode from-sqlite --archive-existing --yes` block), and a decision table mapping needs to tools that covers the common cases. | 5 / 5 |
Workflow Clarity | A numbered 6-step Protocol with clear sequencing is backed by explicit validation checkpoints and feedback loops: a Step-0 availability check, "Stop when answered", the empty-results loop (one empty query does not prove absence — widen scope, then state what was checked rather than inventing an answer), and the corrupt-index recovery path with concrete repair commands and a warning not to re-mine (which drops MCP-added drawers). | 5 / 5 |
Progressive Disclosure | No bundle files exist; the body is self-contained with well-labeled sections and clearly signaled, one-level-deep external links to the canonical protocol docs in the MemPalace repository. Structure is good and navigation is easy, but everything is inlined in a ~136-line body (unhappy-path repair detail and coordination-protocol material could live in separate reference files), so it lands on anchor 4 rather than the well-split anchor 5. | 4 / 5 |
Total | 18 / 20 Passed |