> ## Documentation Index
> Fetch the complete documentation index at: https://filepacks.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Use cases

> Practical ways to use filepacks for evals, agent runs, CI evidence, review handoff, and regression comparison.

filepacks is most useful when a workflow already produces files and you need those files to be easier to trust, review, compare, or hand off.

## Common use cases

| Scenario                           | Input directory               | Why filepacks helps                                                                    | Typical commands            |
| ---------------------------------- | ----------------------------- | -------------------------------------------------------------------------------------- | --------------------------- |
| AI eval result snapshots           | `./eval-output`               | Turns each run into one deterministic artifact you can compare to an accepted baseline | `pack`, `verify`, `compare` |
| Agent run artifacts                | `./agent-output`              | Creates a portable record that another human or agent can inspect later                | `pack`, `inspect`, `verify` |
| Regression comparisons             | `./run-prev`, `./run-current` | Gives a fixed structural diff instead of ad hoc directory comparisons                  | `pack`, `compare`           |
| CI evidence                        | `./build-output`, `./reports` | Preserves the exact generated files from a job for later review                        | `pack`, `verify`, `compare` |
| Prompt or model output comparisons | `./model-a`, `./model-b`      | Compares generated files across prompts, models, or configurations                     | `pack`, `compare`           |
| Human review of generated output   | any generated directory       | Produces one artifact with a canonical manifest and stable digest                      | `pack`, `inspect`, `verify` |
| Portable artifact handoff          | a run output directory        | Makes review independent of the original machine or workspace layout                   | `pack`, `inspect`           |

## Eval snapshot pattern

When each eval run emits files, package each run immediately:

```bash theme={null}
npx filepacks pack ./eval-output --output ./eval-run-42.fpk
npx filepacks verify ./eval-run-42.fpk
npx filepacks compare ./eval-baseline.fpk ./eval-run-42.fpk
```

If `compare` exits `20`, the run changed structurally and needs review.

## Agent run handoff

When an agent finishes producing files:

```bash theme={null}
npx filepacks pack ./agent-output --output ./agent-run-17.fpk
npx filepacks inspect ./agent-run-17.fpk
npx filepacks verify ./agent-run-17.fpk
```

Now the artifact can be uploaded, attached to a ticket, or handed to another reviewer without relying on the original workspace.

## Human review workflow

A simple human review loop is:

1. `inspect` to confirm the artifact identity and scale.
2. `verify` before trusting the artifact.
3. `compare` against the previous accepted artifact.
4. Review the reported changed paths in the original files or your own tooling.

## What filepacks is especially good at

filepacks works best when:

* your workflow already produces files
* you want a deterministic artifact boundary
* you need human-readable review plus automation-friendly exit codes
* you want to preserve evidence locally or in CI without a hosted dependency

## When it is not the right tool

filepacks is not a replacement for:

* Git history and source review
* a registry or artifact store
* semantic eval scoring
* cloud dashboards or workflow orchestration

Use [Why filepacks](/why-filepacks) for the conceptual framing and [Agent workflows](/agent-workflows) for agent-specific patterns.