Content
66%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 well-structured, actionable preset workflow with explicit sequencing, stage guards, and user decision checkpoints, supported by clear one-level reference paths; its main weakness is repeated escalation/delta-spec guidance that inflates token usage without adding clarity.
Suggestions
Consolidate the escalation/upgrade判定 rules into the single '升级判定' section and have other phases reference it once, removing the repeated delta-spec disclaimers from the open and build steps.
Inline the few critical auto-transition/NEXT commands or move them fully to the reference file so the body does not straddle both; pick one home for each instruction.
Specify the on-verify-failure feedback loop in brief (what '验证失败决策' concretely means) rather than only delegating to comet-verify, since batch/CI-affecting changes warrant an explicit retry path.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient task-oriented prose with executable commands, but it repeats the same escalation/upgrade guidance and the delta-spec disclaimer several times across '升级判定', the build phase, and the IMPORTANT blocks, which is padded relative to a leaner treatment — matching 'mostly efficient but includes some unnecessary explanation or could be tightened'. | 3 / 5 |
Actionability | It gives concrete, executable commands (node "$COMET_STATE" init/check/set/transition, node "$COMET_GUARD" ... --apply, openspec status/instructions apply --json) and names exact skills to load, with only minor gaps such as placeholder <name>/<change-name> and reliance on sibling skills for full detail — fitting 'mostly executable guidance; concrete code or commands with minor gaps'. | 4 / 5 |
Workflow Clarity | The four phases (open→build→verify→archive) are clearly sequenced with explicit stage-guard checkpoints, validation requirements (tests/format/commit per task), and pause points for user decisions; it falls just short of a 5 because some feedback loops (e.g., what to do on a verify failure) are delegated to sibling skills rather than fully specified, matching 'clear sequence with most checkpoints present; minor validation gaps'. | 4 / 5 |
Progressive Disclosure | The SKILL.md is an overview that signals one-level-deep references via the comet/reference/*.md paths (scripts.md, context-recovery.md, dirty-worktree.md, debug-gate.md, decision-point.md, auto-transition.md) and delegates detail to sibling skills, with only minor organization gaps such as inline repetition that could live in referenced files — fitting 'good structure; most content is appropriately placed; references mostly clear'. | 4 / 5 |
Total | 15 / 20 Passed |