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.
The body is a lean, disciplined debugging procedure with strong validation gates and concrete escalation rules; it stays well within token budget. Its main weaknesses are the reliance on external plugin files and commands whose contents and behavior are not restated, and a workflow that is narrated rather than explicitly sequenced.
Suggestions
State what the referenced block files and the freeze guard actually enforce in one line each (or vendor them into references/), so the skill remains actionable when the plugin paths are unavailable.
Number the core loop steps explicitly (reproduce → minimize → hypothesize → fix → verify both scenarios → clean up) so the sequence and its checkpoints are unambiguous without reconstruction from prose.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, procedural, and free of explanations of concepts Claude already knows; rules like the HARD-GATE, 3-Strike Rule, and WTF thresholds are all load-bearing. Minor trimmable padding remains in the host-adapter note and license attribution. | 4 / 5 |
Actionability | Mostly executable guidance: a runnable freeze-guard bash snippet, numeric thresholds ('+15% per revert and +20% for touching unrelated files', STOP above 20%), and a copy-ready report format line. It falls short of fully executable because several mechanisms (the contents of skills/blocks/debug-feedback-loop.md, /octo:unfreeze, the freeze guard's enforcement) are referenced but never specified. | 4 / 5 |
Workflow Clarity | A clear sequence is present (reproduce, retain failure signature, minimize, test one named hypothesis, verify minimal repro and original scenario, remove instrumentation, keep a regression test) with an explicit HARD-GATE validation before production changes and feedback loops (3-strike rotation, WTF stop threshold). It is not a 5 because the sequence must be assembled from prose rather than being explicitly ordered end-to-end. | 4 / 5 |
Progressive Disclosure | References are one level deep and clearly signaled with exact paths ('Read skills/blocks/debug-feedback-loop.md from the installed plugin root'), keeping SKILL.md lean. However, no bundle files exist alongside this SKILL.md and all referenced paths point to an assumed external plugin, so navigation depends on unverifiable paths. | 4 / 5 |
Total | 16 / 20 Passed |