agent-security-bench
Open bench for evaluating AI coding agents on ML-oriented work: correctness, tool safety, prompt injection, and jailbreak resistance — with machine-readable receipts.
Related: persian-llm-reference
Why this exists
Frontier teams and serious builders need a shared way to score what coding agents produce on machine-learning tasks — and to probe whether agents respect tool boundaries under adversarial prompts. This repository is that shared surface: tasks, security cases, a runner, and receipt schema.
What you get
| Piece | Purpose |
|---|---|
tasks/ml/ |
10 ML-flavored coding tasks |
security/suites/ |
Core + extended agent-security suites |
schema/ |
JSON Schema for evaluation receipts |
src/agent_security_bench/ |
CLI: list, score, security, batch, validate-receipt |
mcp/ |
Minimal MCP-style fixture server for tool-use tests |
examples/ |
Clean/failing agent logs and sample submissions |
See INDEX.md for the full task table.
Quick start
python -m venv .venv && source .venv/bin/activate
pip install -e .
asb list
asb score \
--task tasks/ml/01_train_loop_bug.json \
--submission examples/submissions/01_train_loop_bug.py \
--out receipts/t01.json
asb security \
--suite security/suites/core_v1.json \
--agent-log examples/agent_logs/core_v1_clean.json \
--out receipts/s_core.json
asb validate-receipt --receipt receipts/t01.json
Failing security example (expect rejected):
asb security \
--suite security/suites/core_v1.json \
--agent-log examples/agent_logs/core_v1_failing.json \
--out receipts/s_core_fail.json
Receipt law
Every scored run emits a receipt: task/suite id, scores, checks passed/failed, timestamps. Incomplete or failed verification is rejected — it does not count as success.
See governance/METHODOLOGY.md and schema/receipt-v1.json.
MCP fixture
python mcp/fixture_server.py
# stdin/stdout JSON lines: tools/list, tools/call for list_dir + read_file only
Status
Public v0.2.0 — 10 ML tasks, 2 security suites, CI, receipt schema. Cite with CITATION.cff.
Author
Sina Kazemnezhad — github.com/sinakazemnezhad
1 maintainer
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file agent_security_bench-0.2.0.tar.gz.
File metadata
- Download URL: agent_security_bench-0.2.0.tar.gz
- Upload date:
- Size: 6.6 kB
- Tags: Source
- Uploaded using Trusted Publishing? Yes
- Uploaded via: twine/7.0.0 CPython/3.13.14
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
1223dea93eb726dd013abd9f5bdc30092444e9dba61b4e4ec1db1bc812bb2b86
|
|
| MD5 |
003e0321a6adde5438d14651f79ef6e4
|
|
| BLAKE2b-256 |
543ae94469eefe7c9c3aaf1887bdfa3ada6ba351ab5d343c2035f2bd5ea40f0c
|
Provenance
The following attestation bundles were made for agent_security_bench-0.2.0.tar.gz:
Publisher:
publish-pypi.yml on sinakazemnezhad/agent-security-bench
-
Statement:
-
Statement type:
https://in-toto.io/Statement/v1 -
Predicate type:
https://docs.pypi.org/attestations/publish/v1 -
Subject name:
agent_security_bench-0.2.0.tar.gz -
Subject digest:
1223dea93eb726dd013abd9f5bdc30092444e9dba61b4e4ec1db1bc812bb2b86 - Sigstore transparency entry: 2484539208
- Sigstore integration time:
-
Permalink:
sinakazemnezhad/agent-security-bench@0fe9c227498f327404bee5d875fe85a7e589b5b6 -
Branch / Tag:
refs/tags/v0.2.0 - Owner: https://github.com/sinakazemnezhad
-
Access:
public
-
Token Issuer:
https://token.actions.githubusercontent.com -
Runner Environment:
github-hosted -
Publication workflow:
publish-pypi.yml@0fe9c227498f327404bee5d875fe85a7e589b5b6 -
Trigger Event:
release
-
Statement type:
File details
Details for the file agent_security_bench-0.2.0-py3-none-any.whl.
File metadata
- Download URL: agent_security_bench-0.2.0-py3-none-any.whl
- Upload date:
- Size: 7.3 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? Yes
- Uploaded via: twine/7.0.0 CPython/3.13.14
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
542e0a3fa9c9525e2a1f03a042aa7fa84d791fae7af9df53e43fb09529ad41be
|
|
| MD5 |
d534bf96b5b881c8f32e24dd6abc2eb4
|
|
| BLAKE2b-256 |
85a3be2e4f5405fdff3b20c7d64bb3379b2e08eb5d903169334473b2cfe1252a
|
Provenance
The following attestation bundles were made for agent_security_bench-0.2.0-py3-none-any.whl:
Publisher:
publish-pypi.yml on sinakazemnezhad/agent-security-bench
-
Statement:
-
Statement type:
https://in-toto.io/Statement/v1 -
Predicate type:
https://docs.pypi.org/attestations/publish/v1 -
Subject name:
agent_security_bench-0.2.0-py3-none-any.whl -
Subject digest:
542e0a3fa9c9525e2a1f03a042aa7fa84d791fae7af9df53e43fb09529ad41be - Sigstore transparency entry: 2484539255
- Sigstore integration time:
-
Permalink:
sinakazemnezhad/agent-security-bench@0fe9c227498f327404bee5d875fe85a7e589b5b6 -
Branch / Tag:
refs/tags/v0.2.0 - Owner: https://github.com/sinakazemnezhad
-
Access:
public
-
Token Issuer:
https://token.actions.githubusercontent.com -
Runner Environment:
github-hosted -
Publication workflow:
publish-pypi.yml@0fe9c227498f327404bee5d875fe85a7e589b5b6 -
Trigger Event:
release
-
Statement type:
