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.
A dense, highly actionable skill body: concrete commands, schemas, thresholds, and validation loops throughout, with the audit and harvest sub-workflows correctly pushed to real reference files. Its main cost is density — long multi-clause paragraphs and inline rule detail (gap taxonomy, ledger parsing) that could be tightened or extracted, plus one raw date outside a deprecated section.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence throughout — no padding, no explanations of known concepts — with dense, information-bearing prose. It misses the level-5 anchor because several 100+ word single-paragraph rule dumps (the seven-gap taxonomy, the ledger-reading rules), a run-on "What is known" step nesting five file-reading instructions, and a bare time-sensitive date "(2026-10-06)" outside any deprecated section could be tightened or split out. | 4 / 5 |
Actionability | Guidance is fully executable: copy-paste-ready commands with argument shapes ("node tooling/scripts/review-ledger.mjs due", "pnpm kb discover <area>"), explicit TSV artifact schemas with column names, a concrete promotion threshold ("impact + novelty + applicability + proofability - risk >= 7"), shard batch sizes, and A–D evidence grades. This matches the level-5 anchor — specific commands and examples covering the common cases. | 5 / 5 |
Workflow Clarity | The pipeline is stated up front ("pick -> question -> registry -> queries -> leads -> read log -> score -> promote") and each phase carries explicit ordering markers ("before the first search", "after every shard", "Before the run ends") plus stop rules. Validation checkpoints and feedback loops are present — "run the helper's check command" after ledger changes, rejected leads with reopen conditions, and per-corpus dispositions with next actions — matching the level-5 anchor. | 5 / 5 |
Progressive Disclosure | The two heavy sub-workflows are properly externalized to real, well-signaled one-level-deep references ([editor audit](./references/editor-audit.md), [test harvest](./references/test-harvest.md)), each linked from the mode that needs it. It sits at level 4 rather than 5 because the main body still inlines substantial detail that could be split out — roughly 30 lines of ledger-parsing rules in "Reading research ledgers" and the seven inline gap-type definitions. | 4 / 5 |
Total | 18 / 20 Passed |