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 exemplary instruction-only skill body: concrete commands, a fail-on-dev validation loop, and zero filler. The only structural improvement is offloading the nine-journey feature map table to a reference file so the core workflow stays lean.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean and repo-specific throughout — every line carries paths, commands, or env vars Claude could not infer ('pnpm world up dev-app-web --detach -- --browser', 'OPENWORK_EVAL_MYSQL_URL=...'). No explanation of known concepts; the one-sentence philosophy ('A change is done when you have watched it work...') earns its place as framing. | 5 / 5 |
Actionability | Fully executable: exact commands for every step (start/teardown, e2e runs, 'git worktree add /tmp/ow-base origin/dev' with the copy/run sequence for the fails-on-dev check), concrete file paths, and per-change-type test locations in a decision table. | 5 / 5 |
Workflow Clarity | Five clearly sequenced steps with an explicit validation checkpoint at step 4 ('must fail' on dev / 'must pass' on branch) and a remediation loop ('If it passes on dev, the assertion is looser than the change: tighten it'). The decision table in step 1 routes change types to the right check. | 5 / 5 |
Progressive Disclosure | Well-organized sections with clearly signaled one-level references to sibling skills ('write-a-spec', 'drive-desktop-cdp', 'open-a-pr'). However, the nine-core-journeys feature map table is reference-grade detail inlined in the main body; moving it to a reference file would keep SKILL.md a tighter overview. Not a 5 because content split is imperfect; not a 3 because navigation and signaling are clean. | 4 / 5 |
Total | 19 / 20 Passed |