Content
67%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.
The body delivers highly actionable, well-validated operational guidance with genuine feedback loops and a well-structured one-level-deep reference bundle. Its main costs are repetition and sheer size: ~485 lines with the depth policy restated three times and long policy blocks (runtime gate, async costs) inlined that would fit better as reference files, and a scattered section order that makes the end-to-end flow harder to follow than its validation quality deserves.
Suggestions
Consolidate the depth-selection policy into a single section and have 'Submit and collect with --async' and 'Configuration' reference it instead of restating it, cutting the most visible repetition.
Move the 'Runtime compatibility gate' and the async 'What it costs' details into files under references/ (like the existing windows-shells/local-triage pattern), leaving a two-line summary and link in SKILL.md.
Reorder sections into a linear flow (prerequisites → depth choice → preflight → submit → collect → assess → act → recover) so 'Quick start' and the execution rules appear before the deep policy material they set up.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body avoids explaining concepts Claude already knows and most tokens carry non-inferable operational policy, but at ~485 lines it is repetitive and could be tightened: the depth-selection policy is restated in three places ('Choose when and how deeply to review', 'Submit and collect with --async', 'Configuration'), the runtime compatibility gate is a dense 20-line run-on block, and push/collect guardrails recur across sections. It sits between anchors 2 and 3 — inefficient in spots but not padded with unnecessary explanations — so anchor 3 with clear tightening opportunities is the best fit. | 3 / 5 |
Actionability | Provides copy-paste-ready shell for the async submit and collect loop (jq guards, mktemp, traps), an exit-code table with explicit actions, a concrete session-context JSON schema, specific error codes (closed_preview, MT-REPO-NO-CREDENTIAL), and exact commands (qodo --version, git merge-base HEAD origin/main). The 'examples are illustrative — confirm with --help' hedge is explicitly justified by CLI version variance, and minor gaps remain in the connected-progress and MCP paths, placing it just below anchor 5's fully-executable-everywhere bar. | 4 / 5 |
Workflow Clarity | Explicit validation checkpoints abound (exit-code table prescribing actions, 'recheck findings against current code', 'Pending, failed, partial or superseded review is not clean') and feedback loops are thorough (fix → verify → re-review; bounded recovery with 'retry once... stop on further failure'). However the sequence is scattered rather than linear: 'Quick start' appears mid-document after the runtime gate, and execution guidance is split across 'Choose execution for the selected review', 'Execution', and 'Submit and collect', requiring the reader to assemble the flow. This lands between anchors 4 and 5 — validation is anchor-5 quality but organization is anchor-4. | 4 / 5 |
Progressive Disclosure | All 8 bundle files in references/ are real, clearly signaled with descriptive link text and stated purpose ([Session context], [field guide], [connected progress], [shell guidance]), and one level deep — they link back to SKILL.md rather than nesting further. The weakness is that the 485-line body inlines substantial material that could live in references (the runtime compatibility gate, the async 'What it costs' details, and connected-progress canonical rules duplicated alongside their own reference file), which is more than the 'minor organization gaps' of anchor 4 but does not reach anchor 3's 'content that should be separate is inline' as the dominant characterization. | 4 / 5 |
Total | 15 / 20 Passed |