Content
85%Weight 40%Scale 1-3Reviews 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, highly actionable multi-phase workflow with strong validation gates and clean progressive disclosure into real reference files. Its only weakness is verbosity — the gating and grounding rationale recurs across phases and could be consolidated.
Suggestions
Consolidate the repeated capability-gating and grounding-philosophy statements (they appear in Phase 0, Phase 1, and the Model Tiers section) into a single canonical statement that later phases reference, to cut token cost.
The 'one rule that is the whole moat' section and the Phase 2 grounding gate overlap in framing the project/external floor; consider merging the conceptual rule into method.md and keeping only the operational gate inline.
Tighten the Phase 4 follow-up prose: the computed-next-step, shape-gate, and selection menus repeat similar 'offer one logical continuation and wait' phrasing and could be expressed once with the per-shape exceptions as a compact table.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Dense and project-specific rather than padded with concept explanation, but the body is long and restates the capability-gating and grounding philosophy across several phases; it could be tightened without losing meaning. | 2 / 3 |
Actionability | Gives concrete executable guidance — a copy-paste scratch-dir bash block, an explicit four-part handoff gate, per-tier scout routing, and named blocking-question primitives per harness — fully actionable rather than descriptive. | 3 / 3 |
Workflow Clarity | A clearly sequenced Phase 0–4 flow with explicit validation checkpoints (the grounding gate, mandatory two-source corroboration at Tier 3, the four-part handoff gate) and feedback loops for error recovery and stalemate. | 3 / 3 |
Progressive Disclosure | The body is an overview that defers detail to one-level-deep, clearly signaled references (method.md, intake.md, boundaries.md, cross-model-panel.md, report.md, and agent personas), all verified to exist as real files with explicit load-triggers. | 3 / 3 |
Total | 11 / 12 Passed |