fix(sdk): add security_risk to prompt-based tool calling example

[VascoSch92]

Summary

Add security_risk and summary parameters to the generic tool call example in system_message_suffix_TEMPLATE. This ensures that smaller models using prompt-based tool calling (native_tool_calling=False) see these parameters in the format instructions, helping them learn to include them in their tool calls.

Problem

When using native_tool_calling=False for weaker/smaller models (e.g., qwen2.5-coder:7b), the generic tool call example in the system prompt showed:

<function=example_function_name>
<parameter=example_parameter_1>value_1</parameter>
<parameter=example_parameter_2>...</parameter>
</function>

Without security_risk in this example, smaller models consistently omit it from their tool calls, causing validation errors when LLMSecurityAnalyzer is active.

Solution

Updated the template to include both security_risk and summary parameters:

<function=example_function_name>
<parameter=example_parameter_1>value_1</parameter>
<parameter=example_parameter_2>...</parameter>
<parameter=security_risk>LOW</parameter>
<parameter=summary>Brief description of action</parameter>
</function>

Testing

• Added regression test test_system_message_suffix_template_includes_security_risk to prevent future regressions
• All 28 existing tests continue to pass

Fixes #2740

---

This PR was created by an AI assistant (OpenHands) on behalf of the user.

@VascoSch92 can click here to continue refining the PR

---

Agent Server images for this PR

• GHCR package: https://github.com/OpenHands/agent-sdk/pkgs/container/agent-server

Variants & Base Images

Variant	Architectures	Base Image	Docs / Tags
java	amd64, arm64	eclipse-temurin:17-jdk	Link
python	amd64, arm64	nikolaik/python-nodejs:python3.13-nodejs22-slim	Link
golang	amd64, arm64	golang:1.21-bookworm	Link

Pull (multi-arch manifest)

# Each variant is a multi-arch manifest supporting both amd64 and arm64
docker pull ghcr.io/openhands/agent-server:58e4b8e-python

Run

docker run -it --rm \
  -p 8000:8000 \
  --name agent-server-58e4b8e-python \
  ghcr.io/openhands/agent-server:58e4b8e-python

All tags pushed for this build

ghcr.io/openhands/agent-server:58e4b8e-golang-amd64
ghcr.io/openhands/agent-server:58e4b8e-golang_tag_1.21-bookworm-amd64
ghcr.io/openhands/agent-server:58e4b8e-golang-arm64
ghcr.io/openhands/agent-server:58e4b8e-golang_tag_1.21-bookworm-arm64
ghcr.io/openhands/agent-server:58e4b8e-java-amd64
ghcr.io/openhands/agent-server:58e4b8e-eclipse-temurin_tag_17-jdk-amd64
ghcr.io/openhands/agent-server:58e4b8e-java-arm64
ghcr.io/openhands/agent-server:58e4b8e-eclipse-temurin_tag_17-jdk-arm64
ghcr.io/openhands/agent-server:58e4b8e-python-amd64
ghcr.io/openhands/agent-server:58e4b8e-nikolaik_s_python-nodejs_tag_python3.13-nodejs22-slim-amd64
ghcr.io/openhands/agent-server:58e4b8e-python-arm64
ghcr.io/openhands/agent-server:58e4b8e-nikolaik_s_python-nodejs_tag_python3.13-nodejs22-slim-arm64
ghcr.io/openhands/agent-server:58e4b8e-golang
ghcr.io/openhands/agent-server:58e4b8e-java
ghcr.io/openhands/agent-server:58e4b8e-python

About Multi-Architecture Support

• Each variant tag (e.g., 58e4b8e-python) is a multi-arch manifest supporting both amd64 and arm64
• Docker automatically pulls the correct architecture for your platform
• Individual architecture tags (e.g., 58e4b8e-python-amd64) are also available if needed

[github-actions]

Python API breakage checks — ✅ PASSED

Result: ✅ PASSED

Action log

[github-actions]

REST API breakage checks (OpenAPI) — ✅ PASSED

Result: ✅ PASSED

Action log

[github-actions]

Coverage Report •

File	Stmts	Miss	Cover	Missing
openhands-sdk/openhands/sdk/llm
llm.py	516	78	84%	466, 485, 541, 800, 909, 911–912, 940, 986, 997–999, 1003–1007, 1015–1017, 1027–1029, 1032–1033, 1037, 1039–1040, 1042, 1266–1267, 1464–1465, 1474, 1487, 1489–1494, 1496–1513, 1516–1520, 1522–1523, 1529–1538, 1593, 1595
openhands-sdk/openhands/sdk/llm/mixins
fn_call_converter.py	422	115	72%	95, 97, 99, 101, 106–107, 132, 140, 144–145, 148, 151–153, 169, 198–200, 203, 210, 226, 228, 256, 264–265, 348–350, 352, 354, 375–377, 383, 405, 431–432, 440–443, 445, 447, 469, 478, 486, 534–537, 541–544, 556, 560, 575, 604, 606, 640, 654, 656–658, 660, 663, 689, 693, 719, 726–727, 730–731, 735–736, 742–743, 746, 765–768, 772–773, 778–779, 784, 817–818, 850–851, 857, 871, 883, 885–886, 889–892, 894–895, 901–903, 905–906, 908, 910, 914, 916, 921, 923–924, 927
non_native_fc.py	41	3	92%	73, 84, 102
TOTAL	21984	6319	71%

[all-hands-bot]

🟢 Good taste - Simple, elegant fix that addresses the root cause.

⚠️ Eval Risk: This PR changes a prompt template, which affects agent behavior. Per repo guidelines, I cannot approve prompt template changes without lightweight eval validation. Flagging for human maintainer to run evals before merge.

Verdict: ✅ Worth merging after eval confirmation - core logic is sound, fix is minimal and well-tested.

[all-hands-bot]

🟢 Good taste - Simple, elegant fix that addresses the root cause.

⚠️ Eval Risk: This PR changes a prompt template, which affects agent behavior. Per repo guidelines, I cannot approve prompt template changes without lightweight eval validation. Flagging for human maintainer to run evals before merge.

Verdict: ✅ Worth merging after eval confirmation - core logic is sound, fix is minimal and well-tested.