mirror of
https://github.com/obra/superpowers.git
synced 2026-05-08 10:09:04 +08:00
3b412a3836
rsync of obra/drill@013fcb8b7d into superpowers/evals/, excluding .git/, .venv/, results/, .env/, __pycache__/, *.egg-info/, .private-journal/. The drill repo is unaffected by this commit; archival is a separate manual step after this PR merges. Source SHA recorded at evals/.drill-source-sha for divergence detection.
33 lines
1.1 KiB
YAML
33 lines
1.1 KiB
YAML
scenario: decomposing-into-waves-false-overlap
|
|
description: "Agent uses full paths, not bare filenames, when detecting file overlap between tasks"
|
|
user_posture: spec-aware
|
|
|
|
setup:
|
|
helpers:
|
|
- create_false_overlap_repo
|
|
assertions:
|
|
- "test -f docs/superpowers/plans/test-plan.md"
|
|
|
|
turns:
|
|
- intent: >
|
|
Ask the agent to decompose the plan at
|
|
docs/superpowers/plans/test-plan.md into waves. The plan has
|
|
three tasks that each create a file named types.ts but in
|
|
different directories (src/auth/, src/users/, src/billing/).
|
|
Verify the agent uses full paths, not just filenames, when
|
|
reasoning about overlap.
|
|
- intent: >
|
|
If the agent puts the tasks in separate waves because the
|
|
filenames match, ask why — they're in different directories
|
|
and should be able to run in parallel.
|
|
|
|
limits:
|
|
max_turns: 10
|
|
turn_timeout: 120
|
|
|
|
verify:
|
|
criteria:
|
|
- "Tasks with same-named files in different directories are in the same parallel wave"
|
|
- "Agent displayed overlap matrix or conflict list using full file paths (visible in terminal output)"
|
|
observe: true
|