Skip to content

Test Harness

agentwood was put through a systematic feature sweep. The tests live in test_check/features/, organized as a tree that mirrors the renders and the documentation.

test_check/features/NN_area[/subgroup.py]   ← the test programs
test_check_renders/NN_area/…                 ← their outputs (+ extracted stills)

The areas

# Area What it verifies
01 Shot modes anchor / location / bare_scene
02 Blocking all 20 formations (2/3/4-person)
03 Cameras static / motion / tracking
04 Lighting 6 presets + hdr + none
05 Animation ~100-anim library, sampled by family
06 Actors movement (walk/move/join) + dialog/TTS
07 Timeline natural end, clip ceiling, set_frame
08 Render engines + fast/production modes
09 Editing the Edit DSL (13 features)
10 Existing tests/ catalogue of the author-written scripts

How a test runs

Most are standalone Blender scripts:

blender --background --python test_check/features/02_blocking/two_person.py

The numbers-only areas (e.g. timeline) write a results.txt (expected vs actual frame counts) that's checked instead of an image.

Start cheap

tests/test_int_solo.py is the fastest sanity check (1 actor, 1 frame, no API key). If it renders, the pipeline is healthy.

Outcome

Areas 01–09 are verified; area 10 catalogues the pre-existing scripts; the LLM area (11) was reviewed but not executed (dependency missing). The sweep surfaced 10 bugs / improvements — see Known Issues.