Shows: setup steps that run in every attempt before the agent starts, and don't count as its
change. setup seeds data/users.json with a script and runs the tests (one step a string, one
{ run, timeout }); the agent writes a script that counts the active users, and the eval runs it
and checks the seeded file isn't in the turn's diff.
Needs: nothing for the fake agent; in this repo, run pnpm install && pnpm build at the root
first. A real agent needs its key in .env.local (ANTHROPIC_API_KEY for Claude Code), Docker,
and pnpm reeval image build --profile claude-code once.
Run:
pnpm evalReal or fake agent: pnpm eval runs a fake agent, with no keys or Docker. It copies the
fake-agent/ fixture into the workspace to mimic what an agent would produce, so the
checks have something to check. pnpm eval --profile claude-code runs the same eval on
Claude Code, and --profile <name> on any other agent in examples/profiles.ts.
Copying it out of the repo: reeval.config.ts imports ../profiles.ts. Copy
profiles.ts into the example and import it from ./profiles.ts, or inline the
one profile you need.
What to look at: each attempt's setup.log, in its directory under .reeval/runs/<run>/. A
step that fails stops the attempt as an infra error, before the agent runs:
$ node scripts/seed.mjs
exit 0 after 25ms
$ node --test
exit 0 after 81ms
TAP version 13
# Subtest: loadUsers reads the seeded users
ok 1 - loadUsers reads the seeded users
…
| File | What it is |
|---|---|
with-setup.eval.ts |
The eval: the environment's setup steps, and two checks |
project/ |
The package the agent starts from, with scripts/seed.mjs |
fake-agent/ |
What the fake agent copies into the workspace |
reeval.config.ts |
The fake agent, and the real agents from ../profiles.ts |
setup and the order an attempt applies an environment in are in the
defineEnvironment reference.