Content
78%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.
A well-structured, highly actionable body with excellent progressive disclosure into verified one-level-deep references and explicit validation gates on the risky apply workflow. Its main weakness is token efficiency: a sizable fraction of the body restates general Terraform best practices Claude already knows, some of it duplicating the reference files it points to.
Suggestions
Compress the textbook sections — Naming Conventions, count vs for_each basics, Variable/Resource Block Ordering, and the Standard Module Structure tree — into one-line rules plus pointers to references/code-patterns.md and references/module-patterns.md, keeping only the Provectus-specific overrides in the body.
De-duplicate the body against the bundle: the count/for_each GOOD/BAD snippets and the naming examples restate content that already lives in code-patterns.md, so the body can carry the rule and the pointer and drop the duplicated code blocks.
Add a short feedback loop to the apply/test workflows (e.g. 'if trivy/checkov fail or the plan shows unexpected destroys: fix and re-run plan before requesting approval') so error recovery is in the body rather than only in quick-reference.md's troubleshooting section.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~465-line body is efficient where it carries the Provectus conventions (pinning, layers, cross-root wiring, MCP resolution, apply workflow), but several sections restate textbook Terraform knowledge Claude already has — naming conventions with six code examples, count vs for_each basics with GOOD/BAD blocks, variable block ordering, the standard module directory tree, and the features-by-version table. This fits anchor 3 ('mostly efficient but includes some unnecessary explanation or could be tightened') rather than anchor 2, because the unnecessary material is interleaved with a large amount of genuinely non-obvious, org-specific signal rather than being pure padding. | 3 / 5 |
Actionability | Fully executable guidance throughout: concrete HCL (locals.tf example, tags merge, variable validation block), exact commands ('terraform plan -out=plan.tfplan', 'trivy config .', 'checkov -d .'), decision matrices, and a four-step apply workflow. The one placeholder ('required_version = "= X.Y.Z"') is explicitly justified — 'Version numbers in this skill are illustrative — never copy them' — with a concrete resolution protocol (MCP lookup, then the Registry HTTP API), which the rubric accepts. | 5 / 5 |
Workflow Clarity | Sequences are clear with explicit checkpoints: the apply workflow numbers steps and gates step 4 on 'Wait for explicit approval', security scanning must 'pass before a plan', and test generation starts with 'Confirm the resource schema'. It falls short of anchor 5 because in-body error-recovery feedback loops are absent — there is no 'if validation/plan fails, fix and retry' loop in the body (troubleshooting is delegated to quick-reference.md), which is a minor validation gap characteristic of anchor 4. The destructive-operation cap does not apply since apply is explicitly gated on approval. | 4 / 5 |
Progressive Disclosure | All seven referenced files (references/aws-stack-layout.md, code-patterns.md, module-patterns.md, testing-frameworks.md, ci-cd-workflows.md, security-compliance.md, quick-reference.md) exist, are one level deep with no nested references, and the deep-link anchors used in the body (e.g. '#security-baseline', '#cost-review', '#stack-layers', '#count-vs-for_each-deep-dive') resolve to real headings. Every detail topic is signaled inline at the point of use and consolidated in a References index with per-file descriptions — clear overview, appropriate split, easy navigation. | 5 / 5 |
Total | 17 / 20 Passed |