Content
80%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 tight, highly actionable instruction-only skill body with concrete tool calls, paths, and content rules. Its main weakness is workflow clarity: merges and rewrites of existing memory files proceed without any verification step despite being destructive batch operations. A minor duplication of the no-rescan instruction could also be trimmed.
Suggestions
Add a validation checkpoint after writing a merged file, e.g. re-read the file with file_reader to confirm the merge preserved existing facts and valid frontmatter before moving to the next item.
Deduplicate the repeated instruction not to list the memories directory (lines 23 and 74), keeping it only in the pre-loaded-files section.
Clarify what counts as 'stale parts' when merging (e.g. superseded facts vs. contradicted preferences) so the destructive drop step is less ambiguous.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and prescriptive with no re-taught concepts, but the "do NOT re-scan the directory" instruction is duplicated ("do NOT re-scan the directory with terminal or file_reader" and "Do NOT use terminal or file_reader to list the memories directory"), which could be trimmed. | 4 / 5 |
Actionability | Fully actionable for an instruction-only skill: exact paths (~/.clacky/memories/<topic-slug>.md), concrete tool calls (file_reader(path: ...)), slug rules with examples, numeric size limits, an absolute-date example, and a complete frontmatter template. | 5 / 5 |
Workflow Clarity | The two-step workflow (pick target file with match/no-match branches, then write) is clearly sequenced, but merging and rewriting existing memory files is destructive and items are processed in batch, with no validation/verification checkpoint — the rubric's destructive/batch cap applies. | 3 / 5 |
Progressive Disclosure | No bundle files exist and none are needed; all ~63 lines of content belong inline in this single-purpose subagent prompt, organized under clear sections (Existing Memory Files, Workflow, Guidelines) with easy navigation. | 5 / 5 |
Total | 17 / 20 Passed |