Content
50%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is reasonably structured and table-driven with a real external reference and some executable state commands, but it is held back by redundant routing sections, placeholder agent-selection templates, and missing validation checkpoints.
Suggestions
Consolidate the repeated tier-routing guidance from "Ecomode Routing Rules", "Agent Selection", and "Delegation Enforcement" into a single table to remove redundancy.
Replace the placeholder "use /prompts:executor for this scoped task" blocks with concrete, copy-paste-ready invocation syntax.
Add validation feedback loops for state lifecycle writes (e.g. verify `omx state read` reflects the intended active value before proceeding).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is table-driven and mostly token-efficient, but routing guidance is restated across "Ecomode Routing Rules", "Agent Selection", and "Delegation Enforcement", which could be tightened. | 2 / 3 |
Actionability | It gives some concrete, executable commands (e.g. `omx state write --input ...`, the config JSON), but the agent-selection blocks are placeholder templates ("use /prompts:executor for this scoped task") rather than real runnable commands. | 2 / 3 |
Workflow Clarity | State-management steps are sequenced (activation/deactivation/cleanup), but there are no validation checkpoints or feedback loops for these lifecycle writes, capping clarity at 2. | 2 / 3 |
Progressive Disclosure | There is one clearly signaled, one-level-deep reference (references/agent-tiers.md, which exists), but the body keeps large routing tables inline that could live in the reference, so organization is only adequate. | 2 / 3 |
Total | 8 / 12 Passed |