# Plan Recoverable Agent Changes with Checkpoints

Agent changes can leave unwanted edits that are difficult to reverse. This guide explains checkpoint workflows for reviewing and restoring project files with a separately obtained implementation.

## Install

```bash
npx skillstore add nexaiguy/agent-checkpoint
```

## Metadata

- Status: approved
- Slug: nexaiguy-agent-checkpoint
- Version: 1.2
- Author version: 1.2
- Skillstore revision: r1
- Version status: invalid
- Tree hash: 07c230d4cfaad958a627650815dbed8e030e460010cc72bc251f70a328f411f9
- Author: nexaiguy
- GitHub username: nexaiguy
- License: MIT
- Repository: https://github.com/nexaiguy/nex-ai-skills/tree/463aaff26593654193deb52f24d18e80ec21e323/skills/agent-checkpoint
- Ref: 3ddfdca559084abffbfdd188f2ba14b597dcfd9c
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network
- Quality score: 78
- Quality tier: bronze
- Public page: https://skillstore.pages.dev/skills/nexaiguy-agent-checkpoint
- Manifest: https://skillstore.pages.dev/api/skills/nexaiguy-agent-checkpoint/manifest

## Capabilities

- Documents checkpoint saving before refactors, dependency upgrades, codemods, and migrations.
- Shows difference summaries and a patch inspection command.
- Illustrates whole-project restoration, dry runs, and restoration of selected files.
- Explains undoing a restore using an automatically saved checkpoint.
- Describes Git and non-Git storage models and capture exclusions.
- Outlines an optional Claude Code hook without supplying its implementation.

## Use Cases

- Prepare a refactor recovery plan: Review how to save and compare checkpoints before changing a payments module.
- Review migration rollback boundaries: Identify excluded files and independent backup needs before a dependency upgrade or migration.
- Evaluate automated checkpoint hooks: Understand the proposed Claude Code hook and list implementation checks before enabling automatic checkpoints.

## Prompt Templates

### Understand the guide

```
Explain the documented checkpoint workflow and identify what this guide-only package cannot execute.
```

### Prepare a checkpoint plan

```
Plan checkpoints for my dependency upgrade. Include capture exclusions, backup requirements, and a check that the separate implementation is installed.
```

### Plan selective recovery

```
Using my supplied difference summary, identify unintended deletions and propose a selective restore dry run. Request approval before changing files.
```

### Assess hook readiness

```
Evaluate the documented Claude Code hook for our workflow. Separate documented claims from verified behavior and list missing implementation evidence.
```

## Limitations

- Only the public guide is included; scripts, extension source, templates, and hook implementation are absent.
- Documented checkpoints remain on the project disk and do not replace independent backups.
- The documented capture excludes ignored files, nested repositories, submodules, and files above the default 50 MB limit.
- The documented runtime needs Python 3.8\+; exact line endings need Git 2.40\+, and Windows symlinks need Developer Mode or administrator rights.

## Best Practices

- Obtain and audit the separate implementation before relying on checkpoint creation or restore guarantees.
- Review capture exclusions and maintain independent backups for important project data.
- Inspect differences and preview restores before approving changes to project files.

## Anti Patterns

- Treating this guide-only package as an installed checkpoint utility.
- Assuming same-disk checkpoints replace backups or include ignored secrets.
- Restoring an entire project without reviewing the target checkpoint and affected files.

## Security Audit

- Audited at: 2026-11-02T12:07:20.594\+00:00
- Summary: All 20 static findings are false positives: Markdown formatting, documented commands, attribution links, and explanatory text do not establish malicious behavior. No evidence found of prompt injection, credential access, exfiltration, or reconnaissance in the supplied guide. The implementation, hooks, and tests are explicitly absent, so their safety claims cannot be verified.

## Stats

- Views: 0
- Downloads: 0
- Favorites: 0
- Popularity score: 0