Use when reviewing Atmos infrastructure repositories, evaluating IaC structure, or assessing components/stacks organization for quality and best practices
From shieldnpx claudepluginhub infraspecdev/tesseract --plugin shieldThis skill uses the workspace's default tool permissions.
red-flags-reference.mdscoring-rubric.mdtemplates.mdObserves Claude Code sessions via hooks to create atomic project-scoped instincts with confidence scores, evolving them into skills, commands, or agents.
Automatically extracts reusable patterns like error resolutions, workarounds, and debugging techniques from Claude Code sessions via Stop hook, saving them as learned skills for reuse.
Provides patterns for continuous autonomous agent loops with loop selection, quality gates, evals, recovery controls, and failure mitigation. Useful for production AI agent workflows.
Structured review of Atmos infrastructure repositories with file-based analysis and implementation planning.
Core principle: Analysis and plans are ALWAYS persisted to files, even on re-reviews. Never skip writing files.
Explore -> Questions (skip if known) -> Evaluate -> Write analysis.md -> Write plan.md
-> Ask User: [proceed | stop | edit plan] -> Review Plan -> Execute Step by Step
Use Glob, Read, and file exploration to understand structure. Check for: atmos.yaml, stacks/, components/terraform/, catalog/, CI/CD config, version constraints, provider/backend files, pre-commit hooks, terraform-docs setup, release tooling, and per-component versioning strategy.
Identify repo type: components-only, stacks-only, or monorepo.
On first review, ask 10-15 questions covering architecture, scale, operations, development, and governance. Skip on re-review if context is known. See templates.md for question categories.
Score 10 dimensions on a 1-5 scale (structure, environments, DRY, naming, layering, security, operability, scalability, CI/CD, blast-radius). See scoring-rubric.md for full rubric and grading scale (A-F).
Write both files to <repo-root>/claude/atmos-repo-review/. Use templates from templates.md. Check findings against red-flags-reference.md.
Present grade summary and offer three choices: proceed, stop, or wait for edits.
Re-read plan.md for user edits, clarify ambiguities, then execute step by step. Announce each step, show changes, update verification checkboxes, and stop on failures.
| Mistake | Why It Matters | What to Do Instead |
|---|---|---|
| Skipping file writes on re-review | Loses audit trail; user expects fresh files each time | Always write both analysis.md and plan.md |
| Auto-executing the plan | User may want to edit plan.md first | Always ask and wait for confirmation |
| Scoring without exploring fully | Leads to inaccurate grades and missed issues | Check all directories, CI config, and gitignore before scoring |
| Ignoring repo type distinction | Components-only repos have different expectations than monorepos | Identify repo type in Phase 1 and adjust criteria accordingly |
| Missing provider/backend nuance | Atmos generates override files at deploy time | Consult red-flags-reference.md for commit rules |