Content
92%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 exceptionally operational skill: every phase has concrete commands, verbatim outputs, explicit validation gates, and per-write confirmation, with conditional loading of the only external reference. The sole trimmable cost is mild redundancy in restating the one-question-per-epic rule.
Suggestions
State the one-ask-per-epic rule once in Phase 3 and reference it from Phase 5 instead of repeating it, saving a few lines without losing the guardrail.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and entirely project-specific with no explanation of concepts Claude already knows, but the ask-one-epic-at-a-time rule is restated three times across Phases 3 and 5 ("Never fold several epics into one question", "A yes covers the epic it answers, and only that one", "This is its own ask, one per epic") and could be consolidated. | 4 / 5 |
Actionability | Everything is executable: verbatim report blocks, exact question wording with option lists, runnable scripts (gdd-structure-check.sh, md-section.sh, adr-status.sh, write-guard.sh), and copy-paste-ready EPIC.md and index templates with placeholder conventions. | 5 / 5 |
Workflow Clarity | Six phases are explicitly sequenced with a copy-and-tick checklist, hard stop conditions ("N is 0: stop"), write validation (write-guard.sh before every write, OVERWRITE handling), and error-recovery feedback loops (a dismissed question is re-asked; an existing EPIC.md triggers an update-in-place ask). | 5 / 5 |
Progressive Disclosure | The bundle's one reference (references/pr-epic-gate.md) is clearly signaled and conditionally loaded ("On a skip, do not load it"), is one level deep, and verified to exist; always-needed workflow content and templates stay inline where they belong. | 5 / 5 |
Total | 19 / 20 Passed |