wujin/dispatch.example.yaml at main · awanawana/wujin · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
# Cairn server API endpoint.
server: "http://cairn-server:8000"

# Shared environment variables merged into every worker.
common_env:
  TSEC_SERVER_HOST: "<SERVER_HOST>:8000"
  TSEC_AGENT_TOKEN: "xxx"

runtime:
  # Scheduler tick interval in seconds.
  interval: 3
  # Global concurrency cap across all workers and all projects.
  max_workers: 16  # [优化] 8→16, 提升全局并发
  # Maximum number of projects that may be actively scheduled at the same time.
  max_running_projects: 4  # [优化] 3→4, 更多工单并发
  # Per-project worker cap. Prevents one project from monopolizing the whole dispatcher.
  max_project_workers: 6  # [优化] 4→6, 单项目更多 intent 并行 explore
  # Timeout for each enabled worker healthcheck.
  healthcheck_timeout: 20
  # "startup_and_task" checks at dispatcher startup and before every task.
  # "startup_only" checks only at startup. "disabled" skips automatic worker checks.
  worker_healthcheck: "startup_only"
  # Prompt set loaded from cairn/src/cairn/dispatcher/prompts/.
  prompt_group: "default"

tasks:
  bootstrap:
    # Main execution timeout for the initial direct solve attempt.
    timeout: 300
    # Time budget for conclude fallback after the main phase times out or returns unusable output.
    conclude_timeout: 90
  reason:
    # Reason decides whether the goal is met and whether new intents should be proposed.
    timeout: 300
    # Hard cap on how many new intents one reason step may create.
    max_intents: 2
  explore:
    # Main execution timeout for exploring a claimed intent.
    timeout: 300
    # Time budget for conclude fallback used to summarize confirmed findings from the same session.
    conclude_timeout: 90

container:
  # Base worker container. The dispatcher execs agent commands inside one container per project.
  image: "ghcr.io/oritera/cairn-worker-container:latest"
  network_mode: "host"
  # "stop" keeps finished containers for inspection; "remove" is cleaner but loses post-run state.
  completed_action: "stop"
  # Add Linux capabilities only when a task truly needs them, such as packet crafting or network admin actions.
  # cap_add:
  #   - "NET_RAW"
  #   - "NET_ADMIN"


workers:
  # Worker priority is ascending: lower numbers are preferred first.
  # max_running is per-worker concurrency. Slower backends such as Pi usually benefit from lower values.
  - name: "claudecode_deepseek-v4-pro"
    type: "claudecode"
    task_types: [bootstrap, reason, explore]
    max_running: 2
    priority: 0
    env:
      ANTHROPIC_MODEL: "deepseek-v4-pro"
      ANTHROPIC_BASE_URL: "https://api.deepseek.com/anthropic"
      ANTHROPIC_AUTH_TOKEN: "sk-xxx"

  - name: "codex_qwen3.6-plus"
    type: "codex"
    task_types: [bootstrap,reason,explore]
    max_running: 2
    priority: 0
    env:
      CODEX_MODEL: "qwen3.6-plus"
      CODEX_BASE_URL: "https://dashscope.aliyuncs.com/compatible-mode/v1"
      OPENAI_API_KEY: "sk-xxx"

  - name: "pi_qwen3.6-plus"
    type: "pi"
    task_types: [bootstrap,reason,explore]
    max_running: 2
    priority: 0
    env:
      PI_MODEL: "qwen3.6-plus"
      PI_BASE_URL: "https://dashscope.aliyuncs.com/compatible-mode/v1"
      PI_API_KEY: "sk-xxx"
      PI_PROVIDER_API: "openai-completions"
      PI_MODEL_CONTEXT_WINDOW: "262144"