Content
67%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 dense, actionable instruction set with concrete commands, paths, and output templates, and the workflow is well-sequenced with explicit error gates. Its main weaknesses are a verbatim ~30-line duplication of the extension-hook logic and placeholder pseudocode in the research-dispatch step.
Suggestions
De-duplicate the extension-hook checking logic: state it once and reference it from the after-plan step, or factor the hook output templates into a reference file — this alone would push conciseness toward 5.
Replace the Phase 0 placeholder dispatch block with a concrete agent-dispatch instruction (tool, role, and expected return format for each research task).
Add a post-generation verification step (e.g., confirm research.md resolves every NEEDS CLARIFICATION before Phase 1, and confirm data-model.md/contracts exist before reporting) to close the workflow's validation loop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The instruction style is lean and assumes Claude's competence, but the ~30-line extension-hook checking block is duplicated verbatim ("Check if `.specify/extensions.yml` exists..." appears fully before the Outline and again under step 5), which is more than a minor trim (anchor 4). It is far from anchor 1-2 padding — no concept explanations — so it sits at anchor 3. | 3 / 5 |
Actionability | Concrete, executable guidance dominates: the exact command "`.specify/scripts/powershell/setup-plan.ps1 -Json`", exact file paths (FEATURE_SPEC, IMPL_PLAN, `.specify/memory/constitution.md`), and exact output templates for hooks. Minor gaps: the Phase 0 research dispatch block ("Task: \"Research {unknown} for {feature context}\"") is placeholder pseudocode rather than a concrete dispatch command. | 4 / 5 |
Workflow Clarity | A clear numbered sequence (setup → load context → plan workflow → stop and report → after-hooks) with explicit error checkpoints ("ERROR if violations unjustified", "ERROR on gate failures or unresolved clarifications") and a Phase 1 prerequisite ("`research.md` complete"). Not a 5 because there is no validate-fix-retry loop over generated artifacts (e.g., re-checking data-model.md against the spec) and the 'Stop and report' step lists what to report but no verification that artifacts exist. | 4 / 5 |
Progressive Disclosure | No bundle files exist (references/, scripts/, assets/ are absent), and the single-file body is well-sectioned (Pre-Execution Checks, Outline, Phases 0-1, Key rules) with project-file references clearly signaled. Minor gap: the two long hook output templates (and their duplicated checking rules) would sit better in a reference file, keeping them inline costs some organization polish versus anchor 5. | 4 / 5 |
Total | 15 / 20 Passed |