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 tight, fully executable multi-step workflow with strong validation feedback loops and a resilience-to-network-failure design. It is lean and actionable; the only soft spot is progressive disclosure, where the long inline body could offload some heuristics to a reference file.
Suggestions
Move the Step 5 changelog-source heuristics (CHANGELOG.md/RELEASES.md/GitHub releases/npm page fallback order) into a references/changelog-sources.md and link to it, shortening the main body.
Consider extracting the drift-detection three-rules block into a reference so Step 2 stays a concise procedure.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean, instruction-dense body with no explanation of concepts Claude already knows; the prose present (peerDependencies as a compatibility contract, overrides as a floor, the alignment target) is genuine non-obvious domain rule, not padding. | 5 / 5 |
Actionability | Fully executable: concrete find/jq pipelines, `bun outdated --filter '*'`, registry WebFetch with the exact fields to take, a verbatim report skeleton, and a classification table with numeric thresholds — copy-paste ready for the common cases. | 5 / 5 |
Workflow Clarity | Steps 0–7 are clearly sequenced with explicit validation checkpoints: a prerequisite probe before starting, a verbatim skeleton written first, a final `grep -c` check (both must print 4), `bun install --frozen-lockfile` verification, and a `bun audit` stop condition. | 5 / 5 |
Progressive Disclosure | Well-organized into clearly headed steps with content appropriately inline (the report template and classification table belong here), and no bundle files exist that should be referenced; minor gap is the length (~130 lines) with no external reference split for the changelog-reading heuristics. | 4 / 5 |
Total | 19 / 20 Passed |