Problem
The mathematical-benchmarks-v1/graph-artifact-composition Harbor task cannot build its verifier image. Its tests/Dockerfile references an input filename that is not present in the task's tests/ build context.
Evidence
The task contains:
tests/agent-workflow-v1-graph-artifact-composition-input.json
environment/input.json
but tests/Dockerfile:5-6 requests:
COPY expected.json mathematical-benchmarks-v1-graph-artifact-composition-input.json test.sh verifier.py verifier_support.py public_contract.json /tests/
COPY mathematical-benchmarks-v1-graph-artifact-composition-input.json /app/input.json
The agent-workflow-v1-... name is the pre-migration filename; the Dockerfile was changed to the mathematical-benchmarks-v1-... name during the taxonomy migration without renaming the file in the task directory.
Reproduction
The all-dataset Harbor Oracle run on revision 528500bf8691982d36d583dc7daf5f68e765cd32 reached this task and failed during verifier environment construction before the agent or verifier ran:
RuntimeError: Docker compose command failed for environment graph-artifact-composition
...
COPY expected.json mathematical-benchmarks-v1-graph-artifact-composition-input.json ... /tests/
ERROR: "/mathematical-benchmarks-v1-graph-artifact-composition-input.json": not found
...
COPY mathematical-benchmarks-v1-graph-artifact-composition-input.json /app/input.json
ERROR: "/mathematical-benchmarks-v1-graph-artifact-composition-input.json": not found
The checked-in task files and git ls-files confirm that no file with the requested mathematical-benchmarks-v1-... name exists. The failure is deterministic and independent of provider availability or model behavior.
Impact
This task is unrunnable in the Oracle and cannot produce a verifier result. It prevents the full mathematical benchmark dataset from completing its validation run.
Suggested fix
Make the two COPY source names match the checked-in agent-workflow-v1-graph-artifact-composition-input.json file, or rename the file and update the task contract consistently. Re-run the Harbor contract checks and the task Oracle after the correction.
Problem
The
mathematical-benchmarks-v1/graph-artifact-compositionHarbor task cannot build its verifier image. Itstests/Dockerfilereferences an input filename that is not present in the task'stests/build context.Evidence
The task contains:
tests/agent-workflow-v1-graph-artifact-composition-input.jsonenvironment/input.jsonbut
tests/Dockerfile:5-6requests:The
agent-workflow-v1-...name is the pre-migration filename; the Dockerfile was changed to themathematical-benchmarks-v1-...name during the taxonomy migration without renaming the file in the task directory.Reproduction
The all-dataset Harbor Oracle run on revision
528500bf8691982d36d583dc7daf5f68e765cd32reached this task and failed during verifier environment construction before the agent or verifier ran:The checked-in task files and
git ls-filesconfirm that no file with the requestedmathematical-benchmarks-v1-...name exists. The failure is deterministic and independent of provider availability or model behavior.Impact
This task is unrunnable in the Oracle and cannot produce a verifier result. It prevents the full mathematical benchmark dataset from completing its validation run.
Suggested fix
Make the two
COPYsource names match the checked-inagent-workflow-v1-graph-artifact-composition-input.jsonfile, or rename the file and update the task contract consistently. Re-run the Harbor contract checks and the task Oracle after the correction.