openobserve/opencode.jsonc at main · endformdev/openobserve · 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
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
{
  // OpenCode configuration for this repo's CI agents.
  // - ai-review-* agents: used by scripts/ai-review/run-review.mjs (dispatched via `opencode serve`)
  //   for PR code review. ONE shared set, run once per provider by the matrix in
  //   .github/workflows/ai-code-review.yml (DeepSeek-V4-Pro). The model is
  //   supplied per-call by the runner, so the agents themselves are provider-agnostic.
  // Docs: https://opencode.ai/docs/  — verify provider shape against current docs at build time.
  "$schema": "https://opencode.ai/config.json",

  "provider": {
    // DeepSeek provider for PR review (DEEPSEEK_API_KEY).
    "deepseek-review": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "DeepSeek (review)",
      "options": {
        "baseURL": "https://api.deepseek.com",
        "apiKey": "{env:DEEPSEEK_API_KEY}"
      },
      "models": {
        "deepseek-v4-pro": {
          "name": "DeepSeek-V4-Pro",
          "limit": { "context": 1048576, "output": 262144 }
        }
      }
    }
  },

  // Default model for runs that don't pass --model explicitly. run-review.mjs sends an explicit
  // model per call, so this is only a fallback for local/manual runs.
  "model": "deepseek-review/deepseek-v4-pro",

  "agent": {
    // PR review agents. Prompts are assembled at call time in run-review.mjs
    // (shared-rules.md + agents/<key>.md) and sent via the `system` override on
    // session.prompt, so no `prompt` is registered here.
    //
    // ONE shared agent set drives every provider. run-review.mjs sends an explicit
    // `model: {providerID, modelID}` (and `variant` when REVIEW_MODEL_VARIANT is set) on each
    // session.prompt call, which overrides whatever is declared here — so the `model` below is
    // only a default for local/manual `opencode` runs. Do NOT fork these into per-provider
    // copies: identical agents are exactly what makes a model-vs-model comparison meaningful,
    // and duplicated blocks drift.
    //
    // Tool policy is an ALLOWLIST-by-negation, not just "no write/edit/bash": PR diff
    // content is untrusted (any external contributor can put prompt-injection text in a
    // diff) and is fed directly into these agents' prompts, with output auto-posted to the
    // PR. Only read/glob/grep are needed to inspect real repo context around the diff.
    // Everything else — especially webfetch/websearch (SSRF / exfiltration via injected
    // instructions) and task (spawning subagents that may not inherit these restrictions)
    // — is explicitly denied. Deny the full known opencode tool-id set (see
    // `opencode serve` + `GET /experimental/tool/ids`) rather than only the obvious ones,
    // so a newly added tool doesn't silently become available by omission.
    "ai-review-security": {
      "mode": "primary",
      "model": "deepseek-review/deepseek-v4-pro",
      "tools": {
        "read": true, "glob": true, "grep": true,
        "write": false, "edit": false, "bash": false, "patch": false, "apply_patch": false,
        "webfetch": false, "websearch": false, "task": false, "todowrite": false, "skill": false
      }
    },
    "ai-review-code-quality": {
      "mode": "primary",
      "model": "deepseek-review/deepseek-v4-pro",
      "tools": {
        "read": true, "glob": true, "grep": true,
        "write": false, "edit": false, "bash": false, "patch": false, "apply_patch": false,
        "webfetch": false, "websearch": false, "task": false, "todowrite": false, "skill": false
      }
    },
    "ai-review-performance": {
      "mode": "primary",
      "model": "deepseek-review/deepseek-v4-pro",
      "tools": {
        "read": true, "glob": true, "grep": true,
        "write": false, "edit": false, "bash": false, "patch": false, "apply_patch": false,
        "webfetch": false, "websearch": false, "task": false, "todowrite": false, "skill": false
      }
    },
    "ai-review-documentation": {
      "mode": "primary",
      "model": "deepseek-review/deepseek-v4-pro",
      "tools": {
        "read": true, "glob": true, "grep": true,
        "write": false, "edit": false, "bash": false, "patch": false, "apply_patch": false,
        "webfetch": false, "websearch": false, "task": false, "todowrite": false, "skill": false
      }
    },
    "ai-review-release": {
      "mode": "primary",
      "model": "deepseek-review/deepseek-v4-pro",
      "tools": {
        "read": true, "glob": true, "grep": true,
        "write": false, "edit": false, "bash": false, "patch": false, "apply_patch": false,
        "webfetch": false, "websearch": false, "task": false, "todowrite": false, "skill": false
      }
    },
    // Reads .claude/skills/ui-architect/references/*.md on demand, so `read`/`glob`/`grep`
    // are load-bearing here rather than incidental — same deny list as the rest.
    "ai-review-frontend": {
      "mode": "primary",
      "model": "deepseek-review/deepseek-v4-pro",
      "tools": {
        "read": true, "glob": true, "grep": true,
        "write": false, "edit": false, "bash": false, "patch": false, "apply_patch": false,
        "webfetch": false, "websearch": false, "task": false, "todowrite": false, "skill": false
      }
    },
    "ai-review-coordinator": {
      "mode": "primary",
      "model": "deepseek-review/deepseek-v4-pro",
      "tools": {
        "read": true, "glob": true, "grep": true,
        "write": false, "edit": false, "bash": false, "patch": false, "apply_patch": false,
        "webfetch": false, "websearch": false, "task": false, "todowrite": false, "skill": false
      }
    }
  }
}