--- title: reeval image build, reeval clean description: Build every sandbox image a run needs, and remove local debug data. --- > ## Documentation Index > Fetch the complete documentation index at: https://docs.reeval.ai/llms.txt > Use this file to discover all available pages before exploring further. `reeval image build` builds the Docker images your evals need before a run. `reeval clean` removes the local debug data runs leave in `.reeval/debug`. ```bash npx reeval image build --dry-run npx reeval image build --profile nightly npx reeval clean --keep 3 ``` ## reeval image build [//]: # (generated:cli-help:image-build) ```text reeval image build [filter...] Build every image the evals' Docker sandboxes need, per harness: reeval's image (bakes in its CLI), from and dockerfile images, install layers. Builds what's missing or stale (a from tag that moved upstream), three at a time, and reports every failure; exit 3 if any image failed --profile Plan with a profile from reeval.config.ts, as reeval run does, so its harness, sandbox and tags pick the images (also REEVAL_PROFILE) --force Rebuild every image, pulling each base again --dry-run List the images, how each is made and how each stands, building nothing (a from tag is looked up in its registry, not pulled) ``` [//]: # (/generated:cli-help:image-build) Filters match as [`reeval run`'s](/reference/cli/run) do: a substring of an eval file's path, an eval's name or a cell. A cell names its harness, so `reeval image build claude-code` builds only the images Claude Code's cells need. A filter that matches nothing exits 2. It collects the evals with `--config`, `--profile` or `REEVAL_PROFILE`, and filters (not `--tag`), then prepares every image, three at a time, the way `reeval run` does before a run. A profile's `defaults.harness` and `defaults.sandbox` pick the images and its `tags` narrow the evals, as they would for the run; an unknown profile exits 2. 1. reeval's image for each harness the plan uses, `reeval-sandbox:-` (for example `cursor-0.1.6`), which bakes that agent's CLI onto `reeval-sandbox-base:latest`: built when it's missing or reeval's base changed. 2. Each `docker({ from })` image: built when missing, and rebuilt when its tag has moved upstream since. 3. Each `docker({ dockerfile })` image and `install` layer: built when its content hash (or its base) is new. `--force` rebuilds all of them and pulls each base again. It keeps going past a failure, then prints a line per image and exits 3 if any failed (2 if the only failures are a missing `docker({ image })`): ```text ✓ docker · cursor: reeval-sandbox:cursor-0.1.0 e04d443bc8e2 ✗ docker:bad.Dockerfile · cursor (1 eval): Building evals/bad.Dockerfile failed: a RUN step exited 42 | broken (full log: .reeval/cache/images/reeval-dockerfile-bad-dockerfile-6d9d1a5c9d335eb6.log) ✓ docker:debian:bookworm-slim · cursor: reeval-sandbox:cursor-0.1.0-debian-bookworm-slim-12f2bc40 41f80062675e … 2 of 4 images failed to build ``` A real `--dry-run` in a project whose evals use several harnesses and images, where `✓` is ready, `+` to build, `~` stale, `✗` missing and `?` unchecked: ```text images 7 sandbox images (nothing built: --dry-run) docker · cursor · 23 evals + base reeval-sandbox:cursor-0.1.6 (to build: not built yet) … docker+0ab0e263 · cursor + install · 1 eval + base reeval-sandbox:cursor-0.1.6 (to build: not built yet) + install 1 command reeval-install: (to build: not built yet) python · cursor · 1 eval + from python:3.12-slim reeval-sandbox:cursor-0.1.6-python-3-12-slim-375f4a61 (to build: on python:3.12-slim (a6e34c598f24)) … ``` Each block names its sandbox and the harness whose image it builds on. An `install` layer's tag hashes its base image's id, so it reads `reeval-install:` until the base is built, and its real tag after. When Docker can't say whether an image is there (the daemon doesn't answer, or `docker` won't start), the step is `?` unchecked, with Docker's error, never "not built yet". `reeval run --dry-run`'s `images` line counts it the same way: ```text docker · claude-code · 1 eval ? base reeval-sandbox:claude-code-1.0.160-patch.b17cde50 (unchecked: docker image inspect reeval-sandbox:claude-code-1.0.160-patch.b17cde50 couldn't run: spawn docker ENOENT) ``` It needs a `reeval.config.*`, as `reeval run` does: with none found it exits 2 (`No reeval.config.ts … here or above. Run reeval init.`). Rebuild after upgrading reeval. ## reeval clean [//]: # (generated:cli-help:clean) ```text reeval clean Remove local debug data (.reeval/debug); --keep keeps the newest n runs --images Also remove the Docker images reeval built that nothing tags any more, and the intermediate -base tags of from builds ``` [//]: # (/generated:cli-help:clean) `reeval run` already keeps only the newest `debug.keep` runs' debug data (default 20). `reeval clean` removes all of it, or all but the newest `--keep ` runs. It never touches `.reeval/runs/` or the history folder. ```text Removed debug data for 4 runs (106.9 KB) from .reeval/debug ``` ## Related - [Sandboxes and toolchains](/guide/sandboxes) - [docker()](/reference/sandbox/docker) - [Debug data](/reference/debug-data)