# Refactor Safely

> >-

- **Type:** Skill
- **Install:** `agentstack add skill-arasz-ai-badger-refactor-safely`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [Arasz](https://agentstack.voostack.com/s/arasz)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [Arasz](https://github.com/Arasz)
- **Source:** https://github.com/Arasz/ai-badger/tree/main/.ai-badger/skills/refactor-safely
- **Website:** https://github.com/Arasz/ai-badger

## Install

```sh
agentstack add skill-arasz-ai-badger-refactor-safely
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Refactor safely

Always preview before applying. The discipline: enumerate every affected location before changing
one, confirm no critical path runs through the target unexamined, then verify with tests after —
never trust that a rename or removal was complete just because it compiled.

The workflow derives from the skill templates the `code-review-graph` project auto-installs
(MIT, © 2026 Tirth Kanani), rewritten here to be tool-agnostic: every step carries a baseline
that needs no graph server.

## Steps

1. **Enumerate every affected location before changing one.**
   Accelerated: a code-graph refactor tool in rename-preview mode (e.g. `refactor_tool` with
   `mode="rename"`; discover the real tool name from the server's own listing) returns the full
   edit list before anything is applied. Baseline: the IDE's rename-refactoring (atomic,
   reference-aware) where available; otherwise `grep -rn` every reference and read each hit to
   rule out false positives (shadowed names, string literals, comments).

2. **Check blast radius and whether a critical path runs through the target.**
   Accelerated: impact-radius / affected-flows calls (e.g. `get_impact_radius`-shaped,
   `get_affected_flows`-shaped). Baseline: grep callers and walk outward toward entry points, same
   as the tracing step in `debug-issue`. Flag any hit in auth, payment, data-write, or public API
   surface.

3. **For a removal, confirm zero *reachable* callers before deleting** — not zero grep hits.
   Accelerated: a dead-code detection mode (e.g. `refactor_tool` with `mode="dead_code"`).
   Baseline: grep the whole tree including test files and dynamic-dispatch registration (DI
   container, route table, reflection lookup, config-driven plugin list) — a zero-hit grep is not
   proof if the language or framework allows dynamic dispatch.

4. **Apply using the safest available mechanism**, always using the edit list from step 1 rather
   than re-deriving it. Accelerated: apply the previewed refactor by its id (e.g.
   `apply_refactor_tool` with the `refactor_id` from step 1). Baseline: the IDE's apply-refactor
   action, or scripted find-replace only after step 1's list is complete and has been read in
   full.

5. **Bracket the apply with test runs** — once before, so a pre-existing failure cannot be
   mistaken for one you caused, and once after, before calling the refactor done. Use the
   configured test command (`.ai-badger/config.json`'s `commands.test`, or the project's
   documented equivalent). Accelerated: a post-refactor change-detection call (e.g.
   `detect_changes`-shaped) to confirm no impact beyond what step 1 previewed. Baseline: run the
   test command directly and read the result — a green exit code from a command you didn't
   actually invoke does not count.

## Gotchas

No environment-specific gotchas known.

## Red flags — STOP

- Applying an edit before step 1 produced a complete list
- "It compiles" treated as proof a rename or removal was complete
- A zero-hit grep treated as proof of dead code, with no check for dynamic dispatch
- An apply with no test run on either side of it

A refactor whose apply is not bracketed by test runs is not a safe refactor, regardless of how
confident the preview looked. Without the run before, a failure after is unattributable.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [Arasz](https://github.com/Arasz)
- **Source:** [Arasz/ai-badger](https://github.com/Arasz/ai-badger)
- **License:** MIT
- **Homepage:** https://github.com/Arasz/ai-badger

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-arasz-ai-badger-refactor-safely
- Seller: https://agentstack.voostack.com/s/arasz
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.