The ageval Core base: what one run is made of
ageval Core architecture: external inputs (User, dataset, profiles) flow through the lock, environment, run, evaluate, and record phases into evidence, with plugin extension points under lock, environment, and run, and limits and cleanup spanning every phase.
INPUTS
LOCK
ENVIRONMENT
RUN
EVALUATE
RECORD
LIMITS
CLEANUP
INPUTS
plugin binding
extension_bindings
capabilities checked here
environment plugins
local · docker · e2b
daytona · ssh
Agent plugins
ACP (default)
nooa · dsh · miniswe
EXT
User
lock / run / view
EXT
dataset
ageval.yaml · tasks/
EXT
profiles
profiles.yaml
CLI
lock
lock.json · digest
PROV
environment
local / docker / e2b…
TASK
run.py
task loop · ACP
EVAL
evaluator.py
PASS / FAIL / ERROR
EVID
evidence
.ageval/runs/<id>/
ageval Core · one run opens one environment; run.py and evaluator.py execute inside it
limits · enforced before the run
(wall-clock · memory · processes · calls)
cleanup · finally, always runs
(every exit path · environment and credentials released)
LEGEND
external · outside the Core
one run · Core boundary
evidence · focal
binds PASS
phase advance
plugin point