Content
88%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.
An unusually dense, fully executable process spec: exact commands, verbatim question texts, output templates, validation guards, and per-write confirmation loops, with the optional producer review properly split into a conditionally-loaded reference file. Only minor tightening is available — some repetition of the ask rules and the fully-inline tier branching keep it just short of maximal token efficiency and ideal file splitting.
Suggestions
State the per-epic ask rule once (e.g., in Phase 3) and reference it from Phase 5 instead of restating it in the checklist, Phase 3, and Phase 5.
Consider moving the tier-specific Phase 2 detail (minimal vs standard/full input loading) into a compact table or short reference to reduce inline branching.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and telegraphic with no padding and no explanation of concepts Claude already knows, but the ask-related rules are defensively restated across the checklist, Phase 3 ("Never fold several epics into one question", "A yes covers the epic it answers, and only that one") and Phase 5 ("This is its own ask, one per epic"), which is minor trimmable repetition. | 4 / 5 |
Actionability | Fully executable throughout: exact commands (e.g., `bash .claude/scripts/md-section.sh extract design/gdd/<file>.md "Overview"`), verbatim report blocks, exact question texts with option lists, complete EPIC.md and index.md markdown templates, and a verbatim closing block covering the common cases. | 5 / 5 |
Workflow Clarity | Six clearly sequenced phases with a copyable checklist, explicit stop conditions ("N is 0: stop"), a write-guard validation before every write with a feedback loop ("an OVERWRITE row back means the file exists after all, so write nothing and ask the update question instead"), and blocked-verdict endings — validation for the batch writes is present, so no destructive-operation cap applies. | 5 / 5 |
Progressive Disclosure | The optional producer review is correctly split into references/pr-epic-gate.md (a real file), one level deep and conditionally loaded ("On a skip, do not load it"), but nearly everything else — three tiers of Phase 2 branching plus both output templates — is inline in a ~250-line body, leaving minor organization gaps below the ideal split. | 4 / 5 |
Total | 18 / 20 Passed |