Content
96%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 high-quality, self-contained skill body: destructive operations are fenced with archive-first and confirmation checkpoints, all three modes (archive, archive+rebuild, restore) have numbered executable steps, and QMD handling includes guards, verification commands, and a failure-report protocol. The only structural note is that the ~200 lines of detail could be split into reference files for lighter context loading.
Suggestions
Consider moving the QMD Refresh section into a references/qmd.md file and keeping a one-line pointer plus the mode table in SKILL.md, so the QMD detail loads only when needed.
The archive directory tree and archive-meta.json example could live in a short references/archive-format.md, leaving SKILL.md as a leaner overview of the three modes.
Clarify in Mode 2 Step 3 how the user invokes the listed re-ingest skills (e.g., whether 'claude-history-ingest' is a sibling skill directory), to remove the only soft spot in an otherwise fully executable workflow.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and instructive throughout: the archive directory tree, archive-meta.json example, log-line templates, and QMD commands all carry operational weight, and no section explains concepts Claude already knows. Minor personality touches ('Archives are cheap insurance', '(obviously)') are single phrases, not padding, so this stays at the top anchor rather than 4. | 5 / 5 |
Actionability | Guidance is concrete and executable: exact archive paths ('_archives/YYYY-MM-DDTHH-MM-SSZ/'), explicit copy targets, ready-to-append log lines, and copy-paste shell commands ('${QMD_CLI:-qmd} update', 'qmd ls "$QMD_WIKI_COLLECTION"', 'qmd get "qmd://..." -l 5'). The enumerated re-ingest skill commands and the QMD-report enum make outcomes deterministic; this matches 'fully executable; copy-paste ready'. | 5 / 5 |
Workflow Clarity | All three modes are clearly sequenced with numbered steps, and the destructive-operation validation requirement is satisfied: 'Always confirm with the user before proceeding', intent confirmation with three modes enumerated, mandatory pre-restore archive, post-restore 'wiki-lint' suggestion, and QMD verification with explicit failure handling ('If QMD refresh fails... report the failure and leave the markdown vault intact'). The destructive-cap at 3 does not apply since validation checkpoints are explicit; feedback loops (verify -> report failure -> leave state intact) are present. | 5 / 5 |
Progressive Disclosure | Sections are well-organized with clear headers, an up-front framing, and appropriate cross-references to the sibling llm-wiki skill's config protocol. However, everything lives inline in a ~200-line SKILL.md with no bundle files — the archive-system layout and the detailed QMD refresh procedure are natural candidates for one-level-deep reference files. This sits between the 4-anchor ('good structure; most content appropriately placed; minor organization gaps') and the 5-anchor; it does not reach 5 because the file carries full operational detail rather than being an overview with well-signaled references. | 4 / 5 |
Total | 19 / 20 Passed |