Content
75%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 well-structured, largely actionable autonomous workflow: concrete rounds with explicit search budgets and a halt condition, full templates for the primary outputs, cross-linking rules, and a closing quality checklist. The main weaknesses are the undefined "standard" concept/entity frontmatter templates, the duplicated research-config instructions, and the absence of an explicit fix-and-re-verify loop after the quality checklist.
Suggestions
Define or point to the 'standard' concept and entity frontmatter templates (or reference the sibling skill/vault convention that defines them) — currently 'Standard concept frontmatter + body' and 'Standard entity frontmatter' leave two of the four filing outputs unspecified.
Merge the 'Research Configuration (optional)' section into Before You Start step 4 — the same vault research-config.md rules are currently stated twice.
Explain how to compute the provenance values (extracted/inferred/ambiguous) and add a brief 'if any checklist item is unchecked, fix it before finishing' loop to the Quality Checklist.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is efficient and assumes Claude's competence — no explaining what web search or Obsidian is — with tight sections like "Run up to 5 targeted searches specifically addressing these gaps". It is not a 5 because of a real redundancy: reading and applying vault `references/research-config.md` is explained twice, once as Before You Start step 4 ("Check $OBSIDIAN_VAULT_PATH/references/research-config.md if it exists") and again as the entire "Research Configuration (optional)" section, which restates the same rules. | 4 / 5 |
Actionability | Guidance is mostly executable: named tools ("WebSearch", "WebFetch", "defuddle <url>"), specific counts ("3-5 distinct angles", "up to 5 targeted searches"), exact paths, full copy-paste YAML frontmatter for reference and synthesis pages, a JSON manifest entry, and a log line format. It is not a 5 because two of the four filing outputs defer to undefined templates — "Standard concept frontmatter + body" and "Standard entity frontmatter" are never shown or pointed to — and the "provenance: extracted: 0.X" placeholders give no semantics for computing the values, so Claude must guess. | 4 / 5 |
Workflow Clarity | The sequence (Before You Start → Round 1 → Round 2 → Round 3 → Filing → Cross-linking → Update Tracking → Quality Checklist) is clear, with an explicit halt condition ("Stop when depth is achieved or 3 rounds are complete"), an ambiguity checkpoint ("Confirm the research topic with the user if it's ambiguous"), and an end-of-run checklist that validates the batch of page writes. It is not a 5 because the checklist lacks a feedback loop — nothing instructs going back and fixing an unchecked item — and there is no verification of written pages' frontmatter correctness; it is above a 3 because validation is present, not absent, for the batch file-writing. | 4 / 5 |
Progressive Disclosure | No bundle files exist (references/, scripts/, assets/ are absent) and the skill is self-contained with clear section headers, well-placed YAML/JSON templates, and a navigable structure, matching the level-4 anchor's "good structure; most content is appropriately placed". It is not a 5 because at ~190 lines the ~75 lines of frontmatter templates could plausibly live in a one-level-deep reference file (the under-50-lines exception for a reference-less 5 does not apply), and the only file mentions (research-config.md, index.md, hot.md) are vault-side runtime paths rather than well-signaled bundle references. | 4 / 5 |
Total | 16 / 20 Passed |