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 highly actionable, well-sequenced skill body: every step gives executable commands, per-agent specifics, and error paths, with a strong feedback loop around the locked tracking-file update. The main weaknesses are the absence of any bundle references (forcing ~300 lines of per-agent detail inline) and a slightly repetitive use-patterns section, plus no post-write validation before the manifest sync.
Suggestions
Move the per-agent tables from Steps 2 and 4 (history paths, index formats, extraction strategies — ~120 lines) into references/agents.md, keeping only the routing and scoring logic in SKILL.md so the body stays a lean overview.
Trim "Cross-Agent Use Patterns" to the 2–3 genuinely distinct scenarios; the Pi examples duplicate the Claude pattern and the no-query example repeats the routing table's default-mode rule.
Add a validation checkpoint after Step 5 (e.g., re-read each written/updated page and verify its frontmatter and cross-links parse before running the manifest sync) so the batch write loop has an explicit verify step.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with tables, commands, and an exact scoring formula, and the one justification it gives ("The 0.35 floor is deliberate: an old session that matches the query exactly must still outrank a recent one that barely matches") earns its tokens. Below 5 because "Cross-Agent Use Patterns" repeats the command-routing table (the two Pi examples duplicate the Claude pattern nearly verbatim) and the QMD section repeats `${QMD_CLI:-qmd}` blocks that could be condensed. | 4 / 5 |
Actionability | Everything is executable: per-agent history paths and config overrides, exact index formats with signal fields, `rg -i "<query terms>"` commands with window sizes (10 before + 20 after), a concrete decay formula, the wiki-page frontmatter template, the locked `obsidian-wiki memory sync` command with arguments, and a fill-in output template. Specific commands cover the common cases for every agent. | 5 / 5 |
Workflow Clarity | A clear 7-step sequence with real checkpoints: missing history root → exact user-facing message; no matches → explicit "No sessions about '<query>' found…" response; "Never hand-edit index.md, log.md, or hot.md — the command takes the lock"; and QMD failure handling ("do not roll back the vault changes; report the QMD status separately"). Below 5 because there is no validation of the written/updated wiki pages before the manifest sync — a batch-write operation whose only pre-write guard is the existing-page check. | 4 / 5 |
Progressive Disclosure | No references/, scripts/, or assets/ files exist in the bundle, so everything is inlined in a ~300-line body — notably the per-agent index tables (Step 2) and extraction strategies (Step 4), roughly 120 lines of agent-specific detail that would fit a single references/agents.md. Section structure and headers are clear and external pointers (llm-wiki/SKILL.md, .skills/llm-wiki/references/MEMORY.md) are named, which lifts it above anchor 2's "minimal structure", but the missing split keeps it below anchor 4's "most content is appropriately placed". | 3 / 5 |
Total | 16 / 20 Passed |