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 well-structured, dense instruction set: modes, triggers, sequencing, stop conditions, output format, and quality gates are all explicit and mostly executable, with a final checklist providing validation. Weaknesses are minor — some repeated policy statements, a few vague directives (outcome-first framing, role consults), and optional-only validation in Direct mode — and no external references are needed or used.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is telegraphic and lean — "Ask one question at a time; gather codebase facts first", "Right-size steps to scope; save to `.omx/plans/`" — with no explanations of concepts Claude already knows. It is not 5 because "`omx explore` is deprecated" is stated twice (Execution_Policy and Tool_Usage) and the 80%/90% thresholds are repeated across Execution_Policy, Final_Checklist, and Advanced, and the "outcome-first framing" sentence is dense filler that could be trimmed. | 4 / 5 |
Actionability | Concrete, executable guidance dominates: explicit flags (`--interview`, `--direct`, `--review`), a save path (`.omx/plans/`), a specific command (`omx question` with `OMX_QUESTION_RETURN_PANE`), verdict values (APPROVED/REVISE/REJECT), and testable thresholds (80%+ claims cite file/line, 90%+ criteria testable). Not 5 because directives like "apply outcome-first framing, concise visible updates" and "consult Analyst for hidden requirements" are non-executable without further definition; not 3 because there is no pseudocode and the specifics cover the common cases. | 4 / 5 |
Workflow Clarity | The multi-step processes are clearly sequenced — a Mode Selection table, numbered steps per mode, and explicit checkpoints ("Stop interviewing when requirements are clear — do not over-interview", the Final Checklist, the Review verdict gate). Not 5 because the Direct mode's validation is only "Optional Critic review if requested" and the interview-to-plan flow lacks an explicit validate-and-revise feedback loop; not 3 because checkpoints are explicit rather than implicit, and the operations are neither destructive nor batch (no cap applies). | 4 / 5 |
Progressive Disclosure | There are no bundle files (references/, scripts/, assets/ are absent) and the body references none, so everything lives inline — appropriately, since the content is compact and policy-like, organized into well-labeled sections (Purpose, Use_When, Do_Not_Use_When, Execution_Policy, Steps, Tool_Usage, Escalation, Final_Checklist, Advanced). Not 5 because at ~90 lines it exceeds the under-50-line simple-skill exception and the Advanced section nests "##" headers inside a tag block, a minor organization gap; not 3 because no content that belongs in separate files is inlined and navigation is easy. | 4 / 5 |
Total | 16 / 20 Passed |