Content
85%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The content is a well-structured, highly actionable orchestration skill with clear sequencing, validation checkpoints, and sound progressive disclosure. Its main weakness is repetition across the decision-point and red-flag sections that inflates token cost.
Suggestions
Consolidate the auto-advance/decision-point guidance: the IMPORTANT block, the 10-node decision-points list, and the Red Flags table restate the same 'do not decide for the user' rule — merge Red Flags into the decision-points list as a single compact reference.
Tighten the build-phase resume rules (build_pause/isolation/build_mode branches) into a small decision table instead of nested prose to reduce repeated qualifiers.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient for a complex state-machine workflow, but the auto-advance/decision-point rules are restated across the IMPORTANT block, the decision-points list, and the Red Flags table, and several resume-rule branches could be tightened without losing clarity. | 2 / 3 |
Actionability | Provides concrete, copy-paste-ready commands (openspec list --json, node "$COMET_INTENT" route --stdin, comet state select, comet guard --apply) plus a complete CometIntentFrame JSON skeleton rather than pseudocode. | 3 / 3 |
Workflow Clarity | Multi-step detection is explicitly sequenced (Step 0 → Step 1 → Step 2 with ordered first-match phase rules) and includes validation checkpoints (phase guards, verify_result feedback loops, an Error Handling quick reference) for risky state transitions. | 3 / 3 |
Progressive Disclosure | Splits a concise Decision Core from a Reference Appendix that points one level deep to clearly signaled comet/reference/*.md files (comet-yaml-fields, file-structure, auto-transition, decision-point, etc.) for detail. | 3 / 3 |
Total | 11 / 12 Passed |