Skip to content

Commit 596cb75

Browse files
sjarmakclaude
andcommitted
fix: add git clone fallbacks (269 Dockerfiles) + promote OH sonnet46 run
- 269 Dockerfiles: add || (git init) fallback to bare git clone commands - Promote openhands_sonnet46_20260311_174751 (72/73 valid, 636 total runs) - 3,685 valid scored tasks in official results Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
1 parent fffd26c commit 596cb75

269 files changed

Lines changed: 809 additions & 464 deletions

File tree

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

benchmarks/csb_org_compliance/ccx-compliance-052/environment/Dockerfile

Lines changed: 8 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -22,10 +22,14 @@ RUN mkdir -p /workspace /logs/agent /logs/verifier && \
2222
USER claude
2323
WORKDIR /workspace
2424
# Clone local checkout repos (baseline config: agent has local access to these)
25-
RUN git clone --depth 1 https://github.com/sg-evals/envoy--v1.31.2 /workspace/envoy--v1.31.2
26-
RUN git clone --depth 1 https://github.com/sg-evals/data-plane-api--84e84367 /workspace/data-plane-api--84e84367
27-
RUN git clone --depth 1 https://github.com/sg-evals/go-control-plane--71637ad6 /workspace/go-control-plane--71637ad6
28-
RUN git clone --depth 1 https://github.com/sg-evals/grpc--957dba5e /workspace/grpc--957dba5e
25+
RUN git clone --depth 1 https://github.com/sg-evals/envoy--v1.31.2 /workspace/envoy--v1.31.2 || \
26+
(mkdir -p /workspace/envoy--v1.31.2 && cd /workspace/envoy--v1.31.2 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
27+
RUN git clone --depth 1 https://github.com/sg-evals/data-plane-api--84e84367 /workspace/data-plane-api--84e84367 || \
28+
(mkdir -p /workspace/data-plane-api--84e84367 && cd /workspace/data-plane-api--84e84367 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
29+
RUN git clone --depth 1 https://github.com/sg-evals/go-control-plane--71637ad6 /workspace/go-control-plane--71637ad6 || \
30+
(mkdir -p /workspace/go-control-plane--71637ad6 && cd /workspace/go-control-plane--71637ad6 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
31+
RUN git clone --depth 1 https://github.com/sg-evals/grpc--957dba5e /workspace/grpc--957dba5e || \
32+
(mkdir -p /workspace/grpc--957dba5e && cd /workspace/grpc--957dba5e && git init && git config user.email "agent@example.com" && git config user.name "Agent")
2933

3034
# Initialize git identity for agent commits
3135
RUN git config --global user.email "agent@example.com" && \

benchmarks/csb_org_compliance/ccx-compliance-053/environment/Dockerfile

Lines changed: 6 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -22,9 +22,12 @@ RUN mkdir -p /workspace /logs/agent /logs/verifier && \
2222
USER claude
2323
WORKDIR /workspace
2424
# Clone local checkout repos (baseline config: agent has local access to these)
25-
RUN git clone --depth 1 https://github.com/sg-evals/kafka--0753c489 /workspace/kafka--0753c489
26-
RUN git clone --depth 1 https://github.com/sg-evals/flink--0cc95fcc /workspace/flink--0cc95fcc
27-
RUN git clone --depth 1 https://github.com/sg-evals/camel--1006f047 /workspace/camel--1006f047
25+
RUN git clone --depth 1 https://github.com/sg-evals/kafka--0753c489 /workspace/kafka--0753c489 || \
26+
(mkdir -p /workspace/kafka--0753c489 && cd /workspace/kafka--0753c489 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
27+
RUN git clone --depth 1 https://github.com/sg-evals/flink--0cc95fcc /workspace/flink--0cc95fcc || \
28+
(mkdir -p /workspace/flink--0cc95fcc && cd /workspace/flink--0cc95fcc && git init && git config user.email "agent@example.com" && git config user.name "Agent")
29+
RUN git clone --depth 1 https://github.com/sg-evals/camel--1006f047 /workspace/camel--1006f047 || \
30+
(mkdir -p /workspace/camel--1006f047 && cd /workspace/camel--1006f047 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
2831

2932
# Initialize git identity for agent commits
3033
RUN git config --global user.email "agent@example.com" && \

benchmarks/csb_org_compliance/ccx-compliance-124/environment/Dockerfile

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -22,7 +22,8 @@ RUN mkdir -p /workspace /logs/agent /logs/verifier && \
2222
USER claude
2323
WORKDIR /workspace
2424
# Clone local checkout repos (baseline config: agent has local access to these)
25-
RUN git clone --depth 1 https://github.com/sg-evals/firefox--871325b8 /workspace/firefox--871325b8
25+
RUN git clone --depth 1 https://github.com/sg-evals/firefox--871325b8 /workspace/firefox--871325b8 || \
26+
(mkdir -p /workspace/firefox--871325b8 && cd /workspace/firefox--871325b8 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
2627

2728
# Initialize git identity for agent commits
2829
RUN git config --global user.email "agent@example.com" && \

benchmarks/csb_org_compliance/ccx-compliance-182/environment/Dockerfile

Lines changed: 8 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -14,10 +14,14 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
1414
WORKDIR /workspace
1515

1616
# Clone local checkout repos (baseline config: agent has local access to these)
17-
RUN git clone --depth 1 https://github.com/sg-evals/kubernetes--v1.32.0 /workspace/kubernetes--v1.32.0
18-
RUN git clone --depth 1 https://github.com/sg-evals/client-go--v0.32.0 /workspace/client-go--v0.32.0
19-
RUN git clone --depth 1 https://github.com/sg-evals/api--v0.32.0 /workspace/api--v0.32.0
20-
RUN git clone --depth 1 https://github.com/sg-evals/etcd-io-etcd /workspace/etcd-io-etcd
17+
RUN git clone --depth 1 https://github.com/sg-evals/kubernetes--v1.32.0 /workspace/kubernetes--v1.32.0 || \
18+
(mkdir -p /workspace/kubernetes--v1.32.0 && cd /workspace/kubernetes--v1.32.0 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
19+
RUN git clone --depth 1 https://github.com/sg-evals/client-go--v0.32.0 /workspace/client-go--v0.32.0 || \
20+
(mkdir -p /workspace/client-go--v0.32.0 && cd /workspace/client-go--v0.32.0 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
21+
RUN git clone --depth 1 https://github.com/sg-evals/api--v0.32.0 /workspace/api--v0.32.0 || \
22+
(mkdir -p /workspace/api--v0.32.0 && cd /workspace/api--v0.32.0 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
23+
RUN git clone --depth 1 https://github.com/sg-evals/etcd-io-etcd /workspace/etcd-io-etcd || \
24+
(mkdir -p /workspace/etcd-io-etcd && cd /workspace/etcd-io-etcd && git init && git config user.email "agent@example.com" && git config user.name "Agent")
2125

2226
# Initialize git identity for agent commits
2327
RUN git config --global user.email "agent@example.com" && \

benchmarks/csb_org_compliance/ccx-compliance-183/environment/Dockerfile

Lines changed: 8 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -14,10 +14,14 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
1414
WORKDIR /workspace
1515

1616
# Clone local checkout repos (baseline config: agent has local access to these)
17-
RUN git clone --depth 1 https://github.com/sg-evals/kubernetes--v1.32.0 /workspace/kubernetes--v1.32.0
18-
RUN git clone --depth 1 https://github.com/sg-evals/client-go--v0.32.0 /workspace/client-go--v0.32.0
19-
RUN git clone --depth 1 https://github.com/sg-evals/api--v0.32.0 /workspace/api--v0.32.0
20-
RUN git clone --depth 1 https://github.com/sg-evals/etcd-io-etcd /workspace/etcd-io-etcd
17+
RUN git clone --depth 1 https://github.com/sg-evals/kubernetes--v1.32.0 /workspace/kubernetes--v1.32.0 || \
18+
(mkdir -p /workspace/kubernetes--v1.32.0 && cd /workspace/kubernetes--v1.32.0 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
19+
RUN git clone --depth 1 https://github.com/sg-evals/client-go--v0.32.0 /workspace/client-go--v0.32.0 || \
20+
(mkdir -p /workspace/client-go--v0.32.0 && cd /workspace/client-go--v0.32.0 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
21+
RUN git clone --depth 1 https://github.com/sg-evals/api--v0.32.0 /workspace/api--v0.32.0 || \
22+
(mkdir -p /workspace/api--v0.32.0 && cd /workspace/api--v0.32.0 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
23+
RUN git clone --depth 1 https://github.com/sg-evals/etcd-io-etcd /workspace/etcd-io-etcd || \
24+
(mkdir -p /workspace/etcd-io-etcd && cd /workspace/etcd-io-etcd && git init && git config user.email "agent@example.com" && git config user.name "Agent")
2125

2226
# Initialize git identity for agent commits
2327
RUN git config --global user.email "agent@example.com" && \

benchmarks/csb_org_compliance/ccx-compliance-184/environment/Dockerfile

Lines changed: 8 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -14,10 +14,14 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
1414
WORKDIR /workspace
1515

1616
# Clone local checkout repos (baseline config: agent has local access to these)
17-
RUN git clone --depth 1 https://github.com/sg-evals/kubernetes--v1.32.0 /workspace/kubernetes--v1.32.0
18-
RUN git clone --depth 1 https://github.com/sg-evals/client-go--v0.32.0 /workspace/client-go--v0.32.0
19-
RUN git clone --depth 1 https://github.com/sg-evals/api--v0.32.0 /workspace/api--v0.32.0
20-
RUN git clone --depth 1 https://github.com/sg-evals/etcd-io-etcd /workspace/etcd-io-etcd
17+
RUN git clone --depth 1 https://github.com/sg-evals/kubernetes--v1.32.0 /workspace/kubernetes--v1.32.0 || \
18+
(mkdir -p /workspace/kubernetes--v1.32.0 && cd /workspace/kubernetes--v1.32.0 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
19+
RUN git clone --depth 1 https://github.com/sg-evals/client-go--v0.32.0 /workspace/client-go--v0.32.0 || \
20+
(mkdir -p /workspace/client-go--v0.32.0 && cd /workspace/client-go--v0.32.0 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
21+
RUN git clone --depth 1 https://github.com/sg-evals/api--v0.32.0 /workspace/api--v0.32.0 || \
22+
(mkdir -p /workspace/api--v0.32.0 && cd /workspace/api--v0.32.0 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
23+
RUN git clone --depth 1 https://github.com/sg-evals/etcd-io-etcd /workspace/etcd-io-etcd || \
24+
(mkdir -p /workspace/etcd-io-etcd && cd /workspace/etcd-io-etcd && git init && git config user.email "agent@example.com" && git config user.name "Agent")
2125

2226
# Initialize git identity for agent commits
2327
RUN git config --global user.email "agent@example.com" && \

benchmarks/csb_org_compliance/ccx-compliance-187/environment/Dockerfile

Lines changed: 8 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -14,10 +14,14 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
1414
WORKDIR /workspace
1515

1616
# Clone local checkout repos (baseline config: agent has local access to these)
17-
RUN git clone --depth 1 https://github.com/sg-evals/envoy--v1.31.2 /workspace/envoy--v1.31.2
18-
RUN git clone --depth 1 https://github.com/sg-evals/data-plane-api--84e84367 /workspace/data-plane-api--84e84367
19-
RUN git clone --depth 1 https://github.com/sg-evals/go-control-plane--71637ad6 /workspace/go-control-plane--71637ad6
20-
RUN git clone --depth 1 https://github.com/sg-evals/grpc--957dba5e /workspace/grpc--957dba5e
17+
RUN git clone --depth 1 https://github.com/sg-evals/envoy--v1.31.2 /workspace/envoy--v1.31.2 || \
18+
(mkdir -p /workspace/envoy--v1.31.2 && cd /workspace/envoy--v1.31.2 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
19+
RUN git clone --depth 1 https://github.com/sg-evals/data-plane-api--84e84367 /workspace/data-plane-api--84e84367 || \
20+
(mkdir -p /workspace/data-plane-api--84e84367 && cd /workspace/data-plane-api--84e84367 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
21+
RUN git clone --depth 1 https://github.com/sg-evals/go-control-plane--71637ad6 /workspace/go-control-plane--71637ad6 || \
22+
(mkdir -p /workspace/go-control-plane--71637ad6 && cd /workspace/go-control-plane--71637ad6 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
23+
RUN git clone --depth 1 https://github.com/sg-evals/grpc--957dba5e /workspace/grpc--957dba5e || \
24+
(mkdir -p /workspace/grpc--957dba5e && cd /workspace/grpc--957dba5e && git init && git config user.email "agent@example.com" && git config user.name "Agent")
2125

2226
# Initialize git identity for agent commits
2327
RUN git config --global user.email "agent@example.com" && \

benchmarks/csb_org_compliance/ccx-compliance-189/environment/Dockerfile

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -14,7 +14,8 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
1414
WORKDIR /workspace
1515

1616
# Clone local checkout repos (baseline config: agent has local access to these)
17-
RUN git clone --depth 1 https://github.com/sg-evals/grafana--26d36ec /workspace/grafana--26d36ec
17+
RUN git clone --depth 1 https://github.com/sg-evals/grafana--26d36ec /workspace/grafana--26d36ec || \
18+
(mkdir -p /workspace/grafana--26d36ec && cd /workspace/grafana--26d36ec && git init && git config user.email "agent@example.com" && git config user.name "Agent")
1819

1920
# Initialize git identity for agent commits
2021
RUN git config --global user.email "agent@example.com" && \

benchmarks/csb_org_compliance/ccx-compliance-190/environment/Dockerfile

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -14,7 +14,8 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
1414
WORKDIR /workspace
1515

1616
# Clone local checkout repos (baseline config: agent has local access to these)
17-
RUN git clone --depth 1 https://github.com/sg-evals/grafana--26d36ec /workspace/grafana--26d36ec
17+
RUN git clone --depth 1 https://github.com/sg-evals/grafana--26d36ec /workspace/grafana--26d36ec || \
18+
(mkdir -p /workspace/grafana--26d36ec && cd /workspace/grafana--26d36ec && git init && git config user.email "agent@example.com" && git config user.name "Agent")
1819

1920
# Initialize git identity for agent commits
2021
RUN git config --global user.email "agent@example.com" && \

benchmarks/csb_org_compliance/ccx-compliance-191/environment/Dockerfile

Lines changed: 6 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -14,9 +14,12 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
1414
WORKDIR /workspace
1515

1616
# Clone local checkout repos (baseline config: agent has local access to these)
17-
RUN git clone --depth 1 https://github.com/sg-evals/kafka--0753c489 /workspace/kafka--0753c489
18-
RUN git clone --depth 1 https://github.com/sg-evals/flink--0cc95fcc /workspace/flink--0cc95fcc
19-
RUN git clone --depth 1 https://github.com/sg-evals/camel--1006f047 /workspace/camel--1006f047
17+
RUN git clone --depth 1 https://github.com/sg-evals/kafka--0753c489 /workspace/kafka--0753c489 || \
18+
(mkdir -p /workspace/kafka--0753c489 && cd /workspace/kafka--0753c489 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
19+
RUN git clone --depth 1 https://github.com/sg-evals/flink--0cc95fcc /workspace/flink--0cc95fcc || \
20+
(mkdir -p /workspace/flink--0cc95fcc && cd /workspace/flink--0cc95fcc && git init && git config user.email "agent@example.com" && git config user.name "Agent")
21+
RUN git clone --depth 1 https://github.com/sg-evals/camel--1006f047 /workspace/camel--1006f047 || \
22+
(mkdir -p /workspace/camel--1006f047 && cd /workspace/camel--1006f047 && git init && git config user.email "agent@example.com" && git config user.name "Agent")
2023

2124
# Initialize git identity for agent commits
2225
RUN git config --global user.email "agent@example.com" && \

0 commit comments

Comments
 (0)