Content
73%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-sequenced, highly actionable orchestration pipeline with explicit validation loops and error handling. Its main cost is redundancy — several rules are restated across <guideline>, <instruct>, <constraint>, and <condition> — which inflates token usage without adding information.
Suggestions
State each rule once (e.g., the 2-iteration halt and never-re-ask rules) in a single canonical section and cross-reference it from <condition> instead of repeating it in <guideline>, <instruct>, and <constraint>.
Reconcile the subagent tooling guidance: R1 prescribes subagent_spawn/subagent_wait with a task_tool caveat, while R2 flatly says "use task_tool" — pick one canonical instruction and apply it to both.
Provide a concrete fill-in template for the R3 review-results output and the R4 transition message instead of bullet-list descriptions.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly tight operational directives, but the 2-iteration halt rule is repeated in <guideline>, <constraint>, and <condition>, the "never re-ask" rule appears three times, and several rationales (test -f portability, free-text confirmation) run long. | 3 / 5 |
Actionability | Concrete tool calls (subagent_spawn/subagent_wait, task_tool, ask_user, sdd_advance, write_file), a copy-paste JSON template, and a full subagent prompt template make most guidance executable; R2's delegation spec and the R3/R4 output formats are described rather than templated. | 4 / 5 |
Workflow Clarity | R1→R4 are clearly sequenced with explicit IF/THEN routing in <condition>, gate feedback loops capped at 2 iterations, error-recovery options, and a decision table for empty/skip/cancel — explicit validation checkpoints throughout. | 5 / 5 |
Progressive Disclosure | A self-contained single file (no bundle files exist) with clear tag- and header-based structure; only minor gaps such as the inlined subagent prompt template in <example> that could live in a reference file. | 4 / 5 |
Total | 16 / 20 Passed |