@kmlckj/licos-ai-cli 1.8.4 → 1.8.5
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/lib/__templates__/agent/config/agent_llm_config.json +3 -1
- package/lib/__templates__/agent/examples/experts.py +7 -4
- package/lib/__templates__/agent/examples/handoff.py +3 -2
- package/lib/__templates__/agent/pyproject.toml +1 -1
- package/lib/__templates__/agent/requirements.txt +1 -1
- package/lib/__templates__/agent/src/agents/agent.py +1 -1
- package/lib/__templates__/workflow/pyproject.toml +1 -1
- package/lib/__templates__/workflow/requirements.txt +1 -1
- package/package.json +1 -1
|
@@ -3,7 +3,7 @@
|
|
|
3
3
|
"model": "auto",
|
|
4
4
|
"temperature": 0.7,
|
|
5
5
|
"top_p": 0.9,
|
|
6
|
-
"max_completion_tokens":
|
|
6
|
+
"max_completion_tokens": 130000,
|
|
7
7
|
"timeout": 600,
|
|
8
8
|
"thinking": "auto"
|
|
9
9
|
},
|
|
@@ -28,6 +28,8 @@
|
|
|
28
28
|
"planning": { "mode": "auto" },
|
|
29
29
|
"context": {
|
|
30
30
|
"strategy": "adaptive",
|
|
31
|
+
"max_context_tokens": 256000,
|
|
32
|
+
"max_input_tokens": 250000,
|
|
31
33
|
"input_budget_ratio": 0.7,
|
|
32
34
|
"compaction_enabled": true,
|
|
33
35
|
"compact_at_ratio": 0.8,
|
|
@@ -22,23 +22,26 @@ def create_agent(ctx):
|
|
|
22
22
|
shared_prompt = "\n\n".join((snapshot.raw["sp"], snapshot.raw["up"]))
|
|
23
23
|
config = snapshot.raw["config"]
|
|
24
24
|
preserve_reasoning = snapshot.harness.thinking.forward_to_frontend
|
|
25
|
+
research_model = configured_model(config, preserve_reasoning=preserve_reasoning)
|
|
26
|
+
review_model = configured_model(config, preserve_reasoning=preserve_reasoning)
|
|
27
|
+
coordinator = configured_model(config, preserve_reasoning=preserve_reasoning)
|
|
25
28
|
endpoint = resolve_llm_endpoint(model=config["model"], model_group=config.get("model_group", "text"))
|
|
26
|
-
harness_run = HarnessRun(snapshot.harness, endpoint.context_length,
|
|
29
|
+
harness_run = HarnessRun(snapshot.harness, endpoint.context_length, coordinator.max_tokens,
|
|
27
30
|
input_limit=endpoint.max_input_length)
|
|
28
31
|
ctx.metadata["agent_harness_run"] = harness_run
|
|
29
32
|
# Each specialist can read its own model configuration from a project file.
|
|
30
33
|
# Use per_thread only when the user wants the specialist to remember prior calls.
|
|
31
34
|
team = Subagents([
|
|
32
35
|
model_subagent("research", "分析问题并整理依据",
|
|
33
|
-
model=
|
|
36
|
+
model=research_model,
|
|
34
37
|
system_prompt=shared_prompt + "\n\n根据提供的资料分析问题。依据不足时明确说明，不编造来源。",
|
|
35
38
|
memory="per_call"),
|
|
36
39
|
model_subagent("review", "检查答案中的逻辑和事实依据",
|
|
37
|
-
model=
|
|
40
|
+
model=review_model,
|
|
38
41
|
system_prompt=shared_prompt + "\n\n审查结论是否得到输入资料支持，并指出不确定性。", memory="per_call"),
|
|
39
42
|
], ctx=ctx)
|
|
40
43
|
return build_project_agent(
|
|
41
|
-
|
|
44
|
+
coordinator, [*TOOLS, *team.tools()], ctx=ctx,
|
|
42
45
|
harness=snapshot.harness, harness_run=harness_run,
|
|
43
46
|
system_prompt=shared_prompt + "\n\n协调研究和审查，按需要调用专家，再向用户提供结论。",
|
|
44
47
|
use_configured_agents=False,
|
|
@@ -14,16 +14,17 @@ def create_agent(ctx):
|
|
|
14
14
|
validate_tool_registry(TOOLS, snapshot.raw["tools"])
|
|
15
15
|
config = snapshot.raw["config"]
|
|
16
16
|
preserve_reasoning = snapshot.harness.thinking.forward_to_frontend
|
|
17
|
+
entry_model = configured_model(config, preserve_reasoning=preserve_reasoning)
|
|
17
18
|
endpoint = resolve_llm_endpoint(model=config["model"], model_group=config.get("model_group", "text"))
|
|
18
19
|
ctx.metadata["agent_harness_run"] = HarnessRun(
|
|
19
|
-
snapshot.harness, endpoint.context_length,
|
|
20
|
+
snapshot.harness, endpoint.context_length, entry_model.max_tokens,
|
|
20
21
|
input_limit=endpoint.max_input_length,
|
|
21
22
|
)
|
|
22
23
|
shared_prompt = "\n\n".join((snapshot.raw["sp"], snapshot.raw["up"]))
|
|
23
24
|
team = Subagents([
|
|
24
25
|
model_handoff_agent(
|
|
25
26
|
"reception", "接待用户并明确需求",
|
|
26
|
-
model=
|
|
27
|
+
model=entry_model, shared_prompt=shared_prompt,
|
|
27
28
|
system_prompt="先了解用户的问题。技术问题交给 support，并说明已确认的信息。",
|
|
28
29
|
destinations={"support": "排查和解释技术问题"}, memory="per_call",
|
|
29
30
|
),
|
|
@@ -40,7 +40,7 @@ def create_agent(_ctx: Any = None) -> Any:
|
|
|
40
40
|
)
|
|
41
41
|
if endpoint.context_length is None:
|
|
42
42
|
raise ValueError(f"Model catalog is missing contextLength for {endpoint.model}")
|
|
43
|
-
reserve =
|
|
43
|
+
reserve = model.max_tokens
|
|
44
44
|
if not isinstance(reserve, int):
|
|
45
45
|
raise ValueError("AGENT Harness needs config.max_completion_tokens for the output reservation")
|
|
46
46
|
harness_run = HarnessRun(
|