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.
An exceptionally well-crafted process skill: lean, imperative prose with zero concept padding, hard phase gates with explicit completion criteria and checklists, and concrete commands and formats at every step. The only real gap is progressive disclosure — the long loop-construction recipe list is inlined where a reference file would keep the core tighter.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body never explains concepts Claude already knows (git bisect, Playwright, RNG seeding, HAR files are all name-checked without tutorial text) and is relentlessly directive — 'Tag every debug log with a unique prefix… Untagged logs survive; tagged logs die.' Every section changes behavior rather than padding context. Not 4 because the few rhetorical moments ('Be aggressive. Be creative. Refuse to give up.') are emphasis that shapes behavior, not over-explanation that could be trimmed. | 5 / 5 |
Actionability | Concrete, executable guidance throughout: 'Tag every debug log with a unique prefix, e.g. [DEBUG-a4f2]', 'a single function call', 'git bisect run it', 'run 1000 random inputs', 'paste the invocation and its output', and the falsifiable hypothesis format 'If <X> is the cause, then <changing Y> will make the bug disappear'. Per the scoring notes, an instruction-only skill with this density of specific commands and formats fully qualifies. Not 4 because there are no pseudocode gaps — every instruction names the tool, format, or command to use. | 5 / 5 |
Workflow Clarity | Six clearly sequenced phases, each with an explicit completion criterion and hard gates between them: 'No red-capable command, no Phase 2', 'Do not proceed until you have reproduced and minimised', and a required pre-done checklist in Phase 6 ('grep the prefix', 're-run the Phase 1 loop'). This matches the top anchor — explicit validation steps, feedback loops, and checklists. Not 4 because checkpoints are not merely present; they block progression and re-verify earlier phases (Phase 5 step 5 re-runs the Phase 1 loop). | 5 / 5 |
Progressive Disclosure | Good structure: clear phase headers, one-level-deep reference to scripts/hitl-loop.template.sh (verified to exist in the bundle), referenced at exactly the point of use. However, at 134 lines the body inlines the 10-item 'Ways to construct one' recipe list — reference-style material that could live in a references/ file so the core SKILL.md stays a lean process overview. Not 5 because content that could be split out is inline; not 3 because what is inline is well-organized and the one reference that exists is clearly signaled. | 4 / 5 |
Total | 19 / 20 Passed |