m4
5a581c78a2
Build / build (push) Has been cancelled
Docker / build (push) Has been cancelled
Lint / ruff (push) Has been cancelled
Test / pytest (ubuntu-latest, 3.11) (push) Has been cancelled
Test / pytest (ubuntu-latest, 3.12) (push) Has been cancelled
Test / pytest (windows-latest, 3.11) (push) Has been cancelled
Test / pytest (windows-latest, 3.12) (push) Has been cancelled
Introduce provider, model, and invocation contracts with encrypted configuration persistence. Add web runtime fencing, route fallback, recovery middleware, workspace scoping, and comprehensive tests.
85 lines
2.6 KiB
Python
85 lines
2.6 KiB
Python
from __future__ import annotations
|
|
|
|
from langchain.agents.middleware.types import ModelRequest
|
|
from langchain_core.messages import HumanMessage
|
|
|
|
from EvoScientist.middleware.skill_context import BudgetedSkillsMiddleware
|
|
|
|
|
|
def _skill(name: str, description: str) -> dict[str, object]:
|
|
return {
|
|
"name": name,
|
|
"description": description,
|
|
"path": f"/skills/{name}/SKILL.md",
|
|
"allowed_tools": [],
|
|
}
|
|
|
|
|
|
def _request(query: str, skills: list[dict[str, object]]) -> ModelRequest:
|
|
return ModelRequest(
|
|
model=object(),
|
|
messages=[HumanMessage(content=query)],
|
|
system_prompt="Base system prompt.",
|
|
state={"skills_metadata": skills},
|
|
)
|
|
|
|
|
|
def test_skill_context_prefers_relevant_skill_and_omits_irrelevant_catalog():
|
|
middleware = BudgetedSkillsMiddleware(
|
|
backend=object(), sources=["/skills/"], max_skills=2, max_skills_bytes=1024
|
|
)
|
|
result = middleware.modify_request(
|
|
_request(
|
|
"请分析蛋白质结构预测结果",
|
|
[
|
|
_skill("protein-structure", "蛋白质结构预测和结果分析。"),
|
|
_skill("frontend-design", "Build polished user interfaces."),
|
|
],
|
|
)
|
|
)
|
|
|
|
prompt = str(result.system_message.content)
|
|
assert "protein-structure" in prompt
|
|
assert "frontend-design" not in prompt
|
|
assert "query-relevant subset" in prompt
|
|
|
|
|
|
def test_skill_context_enforces_count_and_utf8_budget():
|
|
middleware = BudgetedSkillsMiddleware(
|
|
backend=object(),
|
|
sources=["/skills/"],
|
|
max_skills=16,
|
|
max_skills_bytes=2048,
|
|
max_description_bytes=128,
|
|
)
|
|
skills = [
|
|
_skill(f"analysis-{index}", "analysis " + "x" * 1_000) for index in range(300)
|
|
]
|
|
|
|
selected = middleware._select_skills(skills, "analysis")
|
|
rendered = middleware._format_budgeted_skills(selected)
|
|
|
|
assert len(selected) == 16
|
|
assert len(rendered.encode("utf-8")) <= 2048
|
|
assert "analysis-299" not in rendered
|
|
|
|
|
|
def test_skill_context_does_not_fall_back_to_all_skills_without_a_match():
|
|
middleware = BudgetedSkillsMiddleware(backend=object(), sources=["/skills/"])
|
|
result = middleware.modify_request(
|
|
_request(
|
|
"unrelated request",
|
|
[_skill("protein-structure", "Protein folding workflow.")],
|
|
)
|
|
)
|
|
|
|
prompt = str(result.system_message.content)
|
|
assert "protein-structure" not in prompt
|
|
assert "query-relevant subset" in prompt
|
|
|
|
|
|
def test_skill_context_accepts_a_single_skill_source():
|
|
middleware = BudgetedSkillsMiddleware(backend=object(), sources="/skills/")
|
|
|
|
assert middleware.sources == ["/skills/"]
|