About this skill

Problem: complex changes lack a reliable second opinion

When an architecture decision touches three or more modules, when security, permissions, or data-integrity risks are present, or when initial debugging has not resolved the issue, a single executor can easily miss assumptions, boundary conditions, and rollback cost. Allowing an external model to edit files, run commands, or read credentials directly can introduce uncontrolled side effects.

How it works: separate executor from read-only advisor

The skill treats WorkBuddy as the executor and controller: it inspects context, judges risk, applies edits when appropriate, and explains consequences to the user. Claude Code or another reasoning assistant acts only as a read-only advisor, returning architecture analysis, code review, risk notes, assumptions, and recommended next steps. The default semi-auto flow asks for confirmation before sharing minimal relevant context; after advisor input, the executor validates it, separates disagreements, and produces a decision plan. Advisor output is advisory material, not an automatic execution list: no auto-installation, no automatic command execution, no file-write access, and no secret transfer.

Boundaries

Use it for complex refactors, cross-module design, security or compliance review, performance trade-offs, and unresolved debugging. Do not use it for simple edits, formatting, renaming, routine build or test runs, or ordinary git operations. The executor remains accountable for the final recommendation, and high-risk changes should include user confirmation, verification, and rollback preparation.

Screenshots

Use Cases

Before refactoring architecture across three modules, ask a read-only advisor to challenge data-flow assumptions and list risks.
When a security permission change touches credential boundaries, require the advisor to list assumptions and verification items without executing commands.
After the first bug-fix pass misses the root cause, run a structured debug session, compare disagreements, and propose rollback steps.
When performance tuning trades caching against consistency, ask the advisor to state key assumptions and metrics that need validation.

Best For

Backend engineers who want an external model to review large refactors read-only.
Platform engineers handling security, permission, or data-integrity changes.
Tech leads who keep final execution control and do not want models running commands automatically.
Senior engineers who need architecture review risks explained before moving forward.