pantheon-opencode 1.4.1 → 1.4.3
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/README.md +175 -147
- package/bin/pantheon-init.mjs +7 -0
- package/commands/pantheon-model.md +34 -0
- package/package.json +2 -2
- package/scripts/generate-preset-docs.mjs +239 -0
- package/scripts/install/model-picker.mjs +367 -2
- package/scripts/install/opencode.mjs +150 -137
- package/src/agents/zeus.md +14 -25
- package/src/pantheon/cost-command.ts +7 -7
- package/src/pantheon/delegation-enforce.ts +12 -13
- package/src/pantheon/goal-loop.ts +2 -2
- package/src/pantheon/goal-store.ts +1 -1
- package/src/pantheon/hashline/tool.ts +2 -2
- package/src/pantheon/model-command.ts +911 -0
- package/src/pantheon/presets.mjs +30 -33
- package/src/pantheon/tool-context.ts +41 -0
- package/src/plugin.ts +45 -193
- package/src/plugins/tui/dist/tui.d.ts.map +1 -1
- package/src/plugins/tui/dist/tui.js +1 -1
- package/src/plugins/tui/dist/tui.tsx +13 -2
- package/src/plugins/tui/package-lock.json +2 -2
- package/src/plugins/tui/package.json +1 -1
- package/src/plugins/tui/src/index.tsx +21 -8
- package/src/plugins/tui/tsdown.config.ts +11 -0
- package/src/routing.yml +71 -123
- package/src/pantheon/delegation-classifier.ts +0 -307
- package/src/pantheon/delegation-finalize.ts +0 -374
- package/src/pantheon/delegation-notify.ts +0 -197
- package/src/pantheon/delegation-toast.ts +0 -73
- package/src/pantheon/delegation.ts +0 -1110
- package/src/pantheon/permission-eval.ts +0 -284
- package/src/pantheon/step-cap.ts +0 -110
- package/src/pantheon/zeus-delegate-with-retry.ts +0 -401
|
@@ -150,7 +150,7 @@ async function resolvePresetForTui(
|
|
|
150
150
|
}
|
|
151
151
|
|
|
152
152
|
/* ─── Version Detection ────────────────────────────────────
|
|
153
|
-
* 100% dinâmico —
|
|
153
|
+
* 100% dinâmico — fallback embutido em build (__PANTHEON_VERSION__).
|
|
154
154
|
* Se falhar, retorna null e a View omite a versão.
|
|
155
155
|
*
|
|
156
156
|
* Try order:
|
|
@@ -159,7 +159,12 @@ async function resolvePresetForTui(
|
|
|
159
159
|
* 1. api.client.file.read package.json
|
|
160
160
|
* 2. git describe --tags
|
|
161
161
|
* 3. opencode --version
|
|
162
|
-
* 4.
|
|
162
|
+
* 4. __PANTHEON_VERSION__ (tsdown define — build-time embed)
|
|
163
|
+
* 5. null */
|
|
164
|
+
|
|
165
|
+
// Build-time embedded version (tsdown `define` in tsdown.config.ts). Injected
|
|
166
|
+
// only into the bundled dist/tui.js; the raw dist/tui.tsx copy falls through.
|
|
167
|
+
declare const __PANTHEON_VERSION__: string | undefined
|
|
163
168
|
|
|
164
169
|
async function detectVersion(api: TuiPluginApi): Promise<string | null> {
|
|
165
170
|
// Try 0a: installed location (2 levels: dist/tui.js → pantheon-tui/package.json)
|
|
@@ -223,6 +228,12 @@ async function detectVersion(api: TuiPluginApi): Promise<string | null> {
|
|
|
223
228
|
/* fall through */
|
|
224
229
|
}
|
|
225
230
|
|
|
231
|
+
// Try 4: build-time embedded version (tsdown define) — last-resort fallback
|
|
232
|
+
// so the sidebar ALWAYS shows a version even when every probe above fails.
|
|
233
|
+
if (typeof __PANTHEON_VERSION__ !== 'undefined' && __PANTHEON_VERSION__) {
|
|
234
|
+
return __PANTHEON_VERSION__
|
|
235
|
+
}
|
|
236
|
+
|
|
226
237
|
return null // sem fallback — versão não aparece
|
|
227
238
|
}
|
|
228
239
|
|
|
@@ -894,9 +905,10 @@ async function setupUsageBar(api: TuiPluginApi) {
|
|
|
894
905
|
* board row for the same child (a child WITH a report keeps the md
|
|
895
906
|
* entry, source 'md').
|
|
896
907
|
*
|
|
897
|
-
* 2. ENRICHMENT + HISTORY — markdown reports
|
|
898
|
-
* `.pantheon/delegations/<sessionID>/<alias>.md
|
|
899
|
-
* src/pantheon/delegation-finalize.ts
|
|
908
|
+
* 2. ENRICHMENT + HISTORY — markdown reports under
|
|
909
|
+
* `.pantheon/delegations/<sessionID>/<alias>.md`. These are LEGACY files
|
|
910
|
+
* (the writer, src/pantheon/delegation-finalize.ts, was removed); the
|
|
911
|
+
* panel still reads them for history. Each child
|
|
900
912
|
* is matched to its report by the `Task ID` header (== child session
|
|
901
913
|
* id == board task id), which supplies the alias (apo-N), agent,
|
|
902
914
|
* description and terminal duration/timedOut. No report → the child
|
|
@@ -1140,7 +1152,8 @@ export async function readDelegationEntries(dir: string): Promise<DelegationEntr
|
|
|
1140
1152
|
|
|
1141
1153
|
/** Read delegation reports from EVERY session under
|
|
1142
1154
|
* `<root>/.pantheon/delegations/<sessionID>/<alias>.md` — the panel's
|
|
1143
|
-
* HISTORY channel
|
|
1155
|
+
* HISTORY channel (reads legacy files; the writer, delegation-finalize.ts,
|
|
1156
|
+
* was removed). Unlike the children/live channels it does NOT depend on a
|
|
1144
1157
|
* resolved sessionID: with no focused session (null/placeholder), the panel
|
|
1145
1158
|
* still shows the reports from all past sessions (running first, Finalized
|
|
1146
1159
|
* desc — the sort applied by readDelegationEntries). Fail-open: a
|
|
@@ -1149,8 +1162,8 @@ export async function readAllDelegationEntries(root: string): Promise<Delegation
|
|
|
1149
1162
|
return readDelegationEntries(join(root, '.pantheon', 'delegations'))
|
|
1150
1163
|
}
|
|
1151
1164
|
|
|
1152
|
-
/** Resolve the directory
|
|
1153
|
-
* The
|
|
1165
|
+
/** Resolve the directory holding delegation md reports.
|
|
1166
|
+
* The reports live under `.pantheon/delegations` RELATIVE to the server cwd,
|
|
1154
1167
|
* which the TUI exposes as `TuiState.path.directory`. `project` does NOT
|
|
1155
1168
|
* exist on `TuiState.path` (the old `state?.project ?? state?.worktree`
|
|
1156
1169
|
* resolution was always undefined for the first term) and `worktree` is
|
|
@@ -1,6 +1,14 @@
|
|
|
1
|
+
import { readFileSync } from 'node:fs'
|
|
2
|
+
import { join } from 'node:path'
|
|
3
|
+
import { fileURLToPath } from 'node:url'
|
|
1
4
|
import solid from 'rolldown-plugin-solid'
|
|
2
5
|
import { defineConfig } from 'tsdown'
|
|
3
6
|
|
|
7
|
+
const __dirname = fileURLToPath(new URL('.', import.meta.url))
|
|
8
|
+
const pkg = JSON.parse(readFileSync(join(__dirname, 'package.json'), 'utf8')) as {
|
|
9
|
+
version: string
|
|
10
|
+
}
|
|
11
|
+
|
|
4
12
|
export default defineConfig({
|
|
5
13
|
entry: { tui: 'src/index.tsx', server: 'src/server.ts' },
|
|
6
14
|
format: ['esm'],
|
|
@@ -8,6 +16,9 @@ export default defineConfig({
|
|
|
8
16
|
outDir: 'dist',
|
|
9
17
|
clean: true,
|
|
10
18
|
outExtensions: () => ({ js: '.js' }),
|
|
19
|
+
define: {
|
|
20
|
+
__PANTHEON_VERSION__: JSON.stringify(pkg.version),
|
|
21
|
+
},
|
|
11
22
|
plugins: [
|
|
12
23
|
solid({
|
|
13
24
|
solid: {
|
package/src/routing.yml
CHANGED
|
@@ -1,35 +1,31 @@
|
|
|
1
1
|
version: 1.0
|
|
2
2
|
presets:
|
|
3
|
-
go-
|
|
4
|
-
description: "
|
|
3
|
+
go-free:
|
|
4
|
+
description: "Free-tier lineup via OpenCode Zen 2026: big-pickle orchestrator, nemotron-3-super-free planner, qwen3.6-plus-free reviewer, deepseek-v4-flash-free implementers, mimo-v2.5-free scouts. Only -free suffix models; quota, availability and pricing depend on Zen account. 2026 pricing verified."
|
|
5
5
|
vision:
|
|
6
|
-
model: opencode
|
|
7
|
-
reasoning_effort:
|
|
6
|
+
model: opencode/mimo-v2.5-free
|
|
7
|
+
reasoning_effort: low
|
|
8
8
|
providers:
|
|
9
9
|
opencode:
|
|
10
10
|
baseURL: https://opencode.ai/zen/v1
|
|
11
11
|
apiKeyEnv: PANTHEON_OPENCODE_API_KEY
|
|
12
|
-
opencode-go:
|
|
13
|
-
baseURL: https://opencode.ai/zen/go/v1
|
|
14
|
-
apiKeyEnv: PANTHEON_OPENCODE_API_KEY
|
|
15
12
|
agents:
|
|
16
|
-
|
|
17
|
-
|
|
18
|
-
themis: { model: opencode/
|
|
19
|
-
|
|
20
|
-
|
|
21
|
-
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
|
|
26
|
-
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
|
|
30
|
-
talos: { model: opencode/deepseek-v4-flash-free, reasoning_effort: low, fallback_models: [opencode/mimo-v2.5] }
|
|
13
|
+
zeus: { model: opencode/big-pickle, reasoning_effort: medium }
|
|
14
|
+
athena: { model: opencode/nemotron-3-super-free, reasoning_effort: high }
|
|
15
|
+
themis: { model: opencode/qwen3.6-plus-free, reasoning_effort: high }
|
|
16
|
+
hermes: { model: opencode/deepseek-v4-flash-free, reasoning_effort: medium }
|
|
17
|
+
aphrodite: { model: opencode/deepseek-v4-flash-free, reasoning_effort: medium }
|
|
18
|
+
demeter: { model: opencode/deepseek-v4-flash-free, reasoning_effort: medium }
|
|
19
|
+
prometheus: { model: opencode/deepseek-v4-flash-free, reasoning_effort: medium }
|
|
20
|
+
hephaestus: { model: opencode/deepseek-v4-flash-free, reasoning_effort: medium }
|
|
21
|
+
apollo: { model: opencode/mimo-v2.5-free, reasoning_effort: low }
|
|
22
|
+
nyx: { model: opencode/mimo-v2.5-free, reasoning_effort: low }
|
|
23
|
+
gaia: { model: opencode/mimo-v2.5-free, reasoning_effort: low }
|
|
24
|
+
iris: { model: opencode/mimo-v2.5-free, reasoning_effort: low }
|
|
25
|
+
mnemosyne: { model: opencode/mimo-v2.5-free, reasoning_effort: low }
|
|
26
|
+
talos: { model: opencode/mimo-v2.5-free, reasoning_effort: low }
|
|
31
27
|
go-fast:
|
|
32
|
-
description: "
|
|
28
|
+
description: "Low-latency OpenCode Go profile 2026: kimi-k2.7-code and glm-5.3-flash planners, gpt-5.6-luna / mimo-v2.5 / deepseek-v4-flash implementers, gpt-5.6-luna-fast scouts. All via Go gateway https://opencode.ai/zen/go/v1. Pricing 2026."
|
|
33
29
|
vision:
|
|
34
30
|
model: opencode-go/mimo-v2.5
|
|
35
31
|
reasoning_effort: low
|
|
@@ -38,116 +34,68 @@ presets:
|
|
|
38
34
|
baseURL: https://opencode.ai/zen/go/v1
|
|
39
35
|
apiKeyEnv: PANTHEON_OPENCODE_API_KEY
|
|
40
36
|
agents:
|
|
41
|
-
athena: { model: opencode-go/
|
|
42
|
-
themis: { model: opencode-go/
|
|
43
|
-
zeus: { model: opencode-go/
|
|
44
|
-
hermes: { model: opencode-go/deepseek-v4-flash,
|
|
45
|
-
aphrodite: { model: opencode-go/
|
|
46
|
-
demeter: { model: opencode-go/
|
|
47
|
-
prometheus: { model: opencode-go/deepseek-v4-flash,
|
|
48
|
-
hephaestus: { model: opencode-go/
|
|
49
|
-
apollo: { model: opencode-go/
|
|
50
|
-
nyx: { model: opencode-go/
|
|
51
|
-
gaia: { model: opencode-go/
|
|
52
|
-
iris: { model: opencode-go/
|
|
53
|
-
mnemosyne: { model: opencode-go/
|
|
54
|
-
talos: { model: opencode-go/
|
|
55
|
-
go-claude:
|
|
56
|
-
description: "Anthropic Claude tiering: Opus 4.8 for planners/reviewers, Sonnet 5 for implementers, Haiku 4.5 for scouts. Vision turns route to Sonnet 5 (native image support). Prices verified 2026-07."
|
|
57
|
-
vision:
|
|
58
|
-
model: anthropic/claude-sonnet-5
|
|
59
|
-
reasoning_effort: medium
|
|
60
|
-
providers:
|
|
61
|
-
anthropic:
|
|
62
|
-
baseURL: https://api.anthropic.com
|
|
63
|
-
apiKeyEnv: PANTHEON_ANTHROPIC_API_KEY
|
|
64
|
-
agents:
|
|
65
|
-
athena: { model: anthropic/claude-opus-4-8, reasoning_effort: high }
|
|
66
|
-
themis: { model: anthropic/claude-opus-4-8, reasoning_effort: high }
|
|
67
|
-
zeus: { model: anthropic/claude-sonnet-5, reasoning_effort: medium }
|
|
68
|
-
hermes: { model: anthropic/claude-sonnet-5, reasoning_effort: medium }
|
|
69
|
-
aphrodite: { model: anthropic/claude-sonnet-5, reasoning_effort: medium }
|
|
70
|
-
demeter: { model: anthropic/claude-sonnet-5, reasoning_effort: medium }
|
|
71
|
-
prometheus: { model: anthropic/claude-sonnet-5, reasoning_effort: medium }
|
|
72
|
-
hephaestus: { model: anthropic/claude-sonnet-5, reasoning_effort: medium }
|
|
73
|
-
apollo: { model: anthropic/claude-haiku-4-5, reasoning_effort: low }
|
|
74
|
-
nyx: { model: anthropic/claude-haiku-4-5, reasoning_effort: low }
|
|
75
|
-
gaia: { model: anthropic/claude-haiku-4-5, reasoning_effort: low }
|
|
76
|
-
iris: { model: anthropic/claude-haiku-4-5, reasoning_effort: low }
|
|
77
|
-
mnemosyne: { model: anthropic/claude-haiku-4-5, reasoning_effort: low }
|
|
78
|
-
talos: { model: anthropic/claude-haiku-4-5, reasoning_effort: low }
|
|
79
|
-
go-openai:
|
|
80
|
-
description: "OpenAI GPT-5.6 family tiering: Sol for planners/reviewers, Terra for implementers, Luna Fast for scouts. Vision turns route to GPT-5.6 Sol. Prices verified 2026-07."
|
|
81
|
-
vision:
|
|
82
|
-
model: openai/gpt-5.6-sol
|
|
83
|
-
reasoning_effort: high
|
|
84
|
-
providers:
|
|
85
|
-
openai:
|
|
86
|
-
baseURL: https://api.openai.com/v1
|
|
87
|
-
apiKeyEnv: PANTHEON_OPENAI_API_KEY
|
|
88
|
-
agents:
|
|
89
|
-
athena: { model: openai/gpt-5.6-sol, reasoning_effort: high }
|
|
90
|
-
themis: { model: openai/gpt-5.6-sol, reasoning_effort: high }
|
|
91
|
-
zeus: { model: openai/gpt-5.6-terra, reasoning_effort: medium }
|
|
92
|
-
hermes: { model: openai/gpt-5.6-terra, reasoning_effort: medium }
|
|
93
|
-
aphrodite: { model: openai/gpt-5.6-terra, reasoning_effort: medium }
|
|
94
|
-
demeter: { model: openai/gpt-5.6-terra, reasoning_effort: medium }
|
|
95
|
-
prometheus: { model: openai/gpt-5.6-terra, reasoning_effort: medium }
|
|
96
|
-
hephaestus: { model: openai/gpt-5.6-terra, reasoning_effort: medium }
|
|
97
|
-
apollo: { model: openai/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
98
|
-
nyx: { model: openai/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
99
|
-
gaia: { model: openai/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
100
|
-
iris: { model: openai/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
101
|
-
mnemosyne: { model: openai/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
102
|
-
talos: { model: openai/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
37
|
+
athena: { model: opencode-go/kimi-k2.7-code, reasoning_effort: high }
|
|
38
|
+
themis: { model: opencode-go/glm-5.3-flash, reasoning_effort: high }
|
|
39
|
+
zeus: { model: opencode-go/kimi-k2.7-code, reasoning_effort: medium }
|
|
40
|
+
hermes: { model: opencode-go/deepseek-v4-flash, reasoning_effort: medium }
|
|
41
|
+
aphrodite: { model: opencode-go/mimo-v2.5, reasoning_effort: low }
|
|
42
|
+
demeter: { model: opencode-go/gpt-5.6-luna, reasoning_effort: low }
|
|
43
|
+
prometheus: { model: opencode-go/deepseek-v4-flash, reasoning_effort: medium }
|
|
44
|
+
hephaestus: { model: opencode-go/mimo-v2.5, reasoning_effort: low }
|
|
45
|
+
apollo: { model: opencode-go/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
46
|
+
nyx: { model: opencode-go/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
47
|
+
gaia: { model: opencode-go/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
48
|
+
iris: { model: opencode-go/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
49
|
+
mnemosyne: { model: opencode-go/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
50
|
+
talos: { model: opencode-go/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
103
51
|
go-premium:
|
|
104
|
-
description: "
|
|
52
|
+
description: "Premium lineup on OpenCode Go 2026: gpt-5.6-sol / qwen3.8-max / deepseek-v4-pro planners, gpt-5.6-terra / kimi-k3 implementers, gpt-5.6-luna scouts. Vision via gpt-5.6-sol. Pricing 2026."
|
|
105
53
|
vision:
|
|
106
|
-
model: opencode-go/
|
|
107
|
-
reasoning_effort:
|
|
54
|
+
model: opencode-go/gpt-5.6-sol
|
|
55
|
+
reasoning_effort: high
|
|
108
56
|
providers:
|
|
109
57
|
opencode-go:
|
|
110
58
|
baseURL: https://opencode.ai/zen/go/v1
|
|
111
59
|
apiKeyEnv: PANTHEON_OPENCODE_API_KEY
|
|
112
60
|
agents:
|
|
113
|
-
zeus: { model: opencode-go/
|
|
114
|
-
athena: { model: opencode-go/deepseek-v4-pro,
|
|
115
|
-
themis: { model: opencode-go/qwen3.
|
|
116
|
-
hermes: { model: opencode-go/
|
|
117
|
-
aphrodite: { model: opencode-go/kimi-
|
|
118
|
-
demeter: { model: opencode-go/
|
|
119
|
-
prometheus: { model: opencode-go/
|
|
120
|
-
hephaestus: { model: opencode-go/
|
|
121
|
-
apollo: { model: opencode-go/
|
|
122
|
-
nyx: { model: opencode-go/
|
|
123
|
-
gaia: { model: opencode-go/
|
|
124
|
-
iris: { model: opencode-go/
|
|
125
|
-
mnemosyne: { model: opencode-go/
|
|
126
|
-
talos: { model: opencode-go/
|
|
127
|
-
|
|
128
|
-
description: "
|
|
61
|
+
zeus: { model: opencode-go/gpt-5.6-sol, reasoning_effort: medium }
|
|
62
|
+
athena: { model: opencode-go/deepseek-v4-pro, reasoning_effort: high }
|
|
63
|
+
themis: { model: opencode-go/qwen3.8-max, reasoning_effort: high }
|
|
64
|
+
hermes: { model: opencode-go/gpt-5.6-terra, reasoning_effort: medium }
|
|
65
|
+
aphrodite: { model: opencode-go/kimi-k3, reasoning_effort: medium }
|
|
66
|
+
demeter: { model: opencode-go/gpt-5.6-terra, reasoning_effort: medium }
|
|
67
|
+
prometheus: { model: opencode-go/kimi-k3, reasoning_effort: medium }
|
|
68
|
+
hephaestus: { model: opencode-go/gpt-5.6-terra, reasoning_effort: medium }
|
|
69
|
+
apollo: { model: opencode-go/gpt-5.6-luna, reasoning_effort: low }
|
|
70
|
+
nyx: { model: opencode-go/gpt-5.6-luna, reasoning_effort: low }
|
|
71
|
+
gaia: { model: opencode-go/gpt-5.6-luna, reasoning_effort: low }
|
|
72
|
+
iris: { model: opencode-go/gpt-5.6-luna, reasoning_effort: low }
|
|
73
|
+
mnemosyne: { model: opencode-go/gpt-5.6-luna, reasoning_effort: low }
|
|
74
|
+
talos: { model: opencode-go/gpt-5.6-luna, reasoning_effort: low }
|
|
75
|
+
openai:
|
|
76
|
+
description: "Pure OpenAI provider 2026: gpt-5.6-sol planners high, gpt-5.6-luna implementers high, gpt-5.6-luna-fast scouts low (fallback to luna low). Provider openai https://api.openai.com/v1 OPENAI_API_KEY. Pricing 2026."
|
|
129
77
|
vision:
|
|
130
|
-
model:
|
|
131
|
-
reasoning_effort:
|
|
78
|
+
model: openai/gpt-5.6-sol
|
|
79
|
+
reasoning_effort: high
|
|
132
80
|
providers:
|
|
133
|
-
|
|
134
|
-
baseURL: https://
|
|
135
|
-
apiKeyEnv:
|
|
81
|
+
openai:
|
|
82
|
+
baseURL: https://api.openai.com/v1
|
|
83
|
+
apiKeyEnv: OPENAI_API_KEY
|
|
136
84
|
agents:
|
|
137
|
-
|
|
138
|
-
|
|
139
|
-
|
|
140
|
-
hermes: { model:
|
|
141
|
-
aphrodite: { model:
|
|
142
|
-
demeter: { model:
|
|
143
|
-
prometheus: { model:
|
|
144
|
-
hephaestus: { model:
|
|
145
|
-
apollo: { model:
|
|
146
|
-
nyx: { model:
|
|
147
|
-
gaia: { model:
|
|
148
|
-
iris: { model:
|
|
149
|
-
mnemosyne: { model:
|
|
150
|
-
talos: { model:
|
|
85
|
+
athena: { model: openai/gpt-5.6-sol, reasoning_effort: high }
|
|
86
|
+
themis: { model: openai/gpt-5.6-sol, reasoning_effort: high }
|
|
87
|
+
zeus: { model: openai/gpt-5.6-sol, reasoning_effort: high }
|
|
88
|
+
hermes: { model: openai/gpt-5.6-luna, reasoning_effort: high }
|
|
89
|
+
aphrodite: { model: openai/gpt-5.6-luna, reasoning_effort: high }
|
|
90
|
+
demeter: { model: openai/gpt-5.6-luna, reasoning_effort: high }
|
|
91
|
+
prometheus: { model: openai/gpt-5.6-luna, reasoning_effort: high }
|
|
92
|
+
hephaestus: { model: openai/gpt-5.6-luna, reasoning_effort: high }
|
|
93
|
+
apollo: { model: openai/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
94
|
+
nyx: { model: openai/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
95
|
+
gaia: { model: openai/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
96
|
+
iris: { model: openai/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
97
|
+
mnemosyne: { model: openai/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
98
|
+
talos: { model: openai/gpt-5.6-luna-fast, reasoning_effort: low }
|
|
151
99
|
agents:
|
|
152
100
|
zeus:
|
|
153
101
|
role: Central orchestrator
|
|
@@ -1,307 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Delegation Classifier — stuck-agent detection for delegation results.
|
|
3
|
-
*
|
|
4
|
-
* Classifies delegation results into actionable statuses:
|
|
5
|
-
* - `content` / `success` — agent returned usable output
|
|
6
|
-
* - `empty` (empty-mode1 / empty-mode2) — no or near-no output; retryable
|
|
7
|
-
* - `budget-exhausted` — agent hit max_steps; partial result available
|
|
8
|
-
* - `timeout` (timeout-with-partial / timeout-empty / timeout-exhausted)
|
|
9
|
-
* - `error` — provider or infrastructure failure
|
|
10
|
-
*
|
|
11
|
-
* Integrates with the existing `dispatch-guard.ts` classification for
|
|
12
|
-
* empty-response detection and provides a structured `DelegationResult`
|
|
13
|
-
* that replaces raw-text returns.
|
|
14
|
-
*
|
|
15
|
-
* @module delegation-classifier
|
|
16
|
-
*/
|
|
17
|
-
|
|
18
|
-
// ─── Types ─────────────────────────────────────────────────────────────
|
|
19
|
-
|
|
20
|
-
/** Empty-response mode classification (matches dispatch-guard). */
|
|
21
|
-
export type EmptyMode = 'empty-mode1' | 'empty-mode2'
|
|
22
|
-
|
|
23
|
-
/** Timeout sub-type based on partial output and retry history. */
|
|
24
|
-
export type TimeoutSubType = 'timeout-with-partial' | 'timeout-empty' | 'timeout-exhausted'
|
|
25
|
-
|
|
26
|
-
/** Structured delegation result replacing raw-text returns. */
|
|
27
|
-
export interface DelegationResult {
|
|
28
|
-
status:
|
|
29
|
-
| 'success'
|
|
30
|
-
| 'empty'
|
|
31
|
-
| 'budget-exhausted'
|
|
32
|
-
| 'timeout'
|
|
33
|
-
| 'error'
|
|
34
|
-
| 'startup_failed'
|
|
35
|
-
| 'bootstrap_unknown'
|
|
36
|
-
content: string
|
|
37
|
-
retryCount: number
|
|
38
|
-
/** When status is 'empty', which empty mode was detected. */
|
|
39
|
-
classification?: EmptyMode | undefined
|
|
40
|
-
/** When status is 'timeout', the timeout sub-type. */
|
|
41
|
-
subType?: TimeoutSubType | undefined
|
|
42
|
-
/** Available when budget-exhausted or timeout-with-partial. */
|
|
43
|
-
partialResult?: string | undefined
|
|
44
|
-
/** Actionable recommendation for the orchestrator. */
|
|
45
|
-
recommendation?: string | undefined
|
|
46
|
-
}
|
|
47
|
-
|
|
48
|
-
/** Result of empty-response classification. */
|
|
49
|
-
export interface EmptyClassification {
|
|
50
|
-
mode: EmptyMode | 'content'
|
|
51
|
-
hasContent: boolean
|
|
52
|
-
hasTokens: boolean
|
|
53
|
-
}
|
|
54
|
-
|
|
55
|
-
/** Timeout classification input. */
|
|
56
|
-
export interface TimeoutClassifyInput {
|
|
57
|
-
/** Report markdown (undefined if no report was written). */
|
|
58
|
-
report: string | undefined
|
|
59
|
-
/** Whether the child session had any messages at all. */
|
|
60
|
-
hasMessages: boolean
|
|
61
|
-
/** How many retries have already been attempted (0 or 1). */
|
|
62
|
-
retryCount: number
|
|
63
|
-
}
|
|
64
|
-
|
|
65
|
-
/** Result of stuck-agent classification. */
|
|
66
|
-
export interface StuckClassification {
|
|
67
|
-
status: 'success' | 'budget-exhausted' | 'empty'
|
|
68
|
-
partialResult?: string | undefined
|
|
69
|
-
recommendation?: string | undefined
|
|
70
|
-
}
|
|
71
|
-
|
|
72
|
-
// ─── Patterns ──────────────────────────────────────────────────────────
|
|
73
|
-
|
|
74
|
-
/** Matches budget-exhausted messages from the agent or step-cap system. */
|
|
75
|
-
const BUDGET_EXHAUSTED_RE =
|
|
76
|
-
/maximum\s+(?:steps?|step\s*budget)\s+(?:for\s+this\s+agent\s+)?(?:has|have)\s+been\s+reached|step\s*budget\s+(?:was\s+)?exhausted/i
|
|
77
|
-
|
|
78
|
-
/** Output-captured marker in delegation reports. */
|
|
79
|
-
const NO_OUTPUT_RE = /_No output captured\._/i
|
|
80
|
-
|
|
81
|
-
// ─── Empty-Response Classification ────────────────────────────────────
|
|
82
|
-
|
|
83
|
-
/**
|
|
84
|
-
* Classify whether a response is empty, empty-with-tokens, or has content.
|
|
85
|
-
* Matches the dispatch-guard classification (empty-mode1, empty-mode2).
|
|
86
|
-
*
|
|
87
|
-
* @param text The response text (may be null/undefined).
|
|
88
|
-
* @param tokens Optional token counts for mode2 detection.
|
|
89
|
-
*/
|
|
90
|
-
export function classifyEmptyResponse(
|
|
91
|
-
text: string | null | undefined,
|
|
92
|
-
tokens?: { tokensInput?: number; tokensOutput?: number },
|
|
93
|
-
): EmptyClassification {
|
|
94
|
-
const hasContent = typeof text === 'string' && text.trim() !== ''
|
|
95
|
-
if (hasContent) {
|
|
96
|
-
return { mode: 'content', hasContent: true, hasTokens: hasTokens(tokens) }
|
|
97
|
-
}
|
|
98
|
-
const hasT = hasTokens(tokens)
|
|
99
|
-
return {
|
|
100
|
-
mode: hasT ? 'empty-mode2' : 'empty-mode1',
|
|
101
|
-
hasContent: false,
|
|
102
|
-
hasTokens: hasT,
|
|
103
|
-
}
|
|
104
|
-
}
|
|
105
|
-
|
|
106
|
-
/** Whether token counts indicate the model actually ran. */
|
|
107
|
-
function hasTokens(tokens?: { tokensInput?: number; tokensOutput?: number }): boolean {
|
|
108
|
-
if (tokens === undefined) return false
|
|
109
|
-
return (tokens.tokensInput ?? 0) > 0 || (tokens.tokensOutput ?? 0) > 0
|
|
110
|
-
}
|
|
111
|
-
|
|
112
|
-
// ─── Stuck-Agent Detection ────────────────────────────────────────────
|
|
113
|
-
|
|
114
|
-
/**
|
|
115
|
-
* Classify a report as success, budget-exhausted, or empty.
|
|
116
|
-
* Detects max-steps patterns in the report text.
|
|
117
|
-
*
|
|
118
|
-
* @param report The delegation report markdown.
|
|
119
|
-
*/
|
|
120
|
-
export function classifyStuckAgent(report: string): StuckClassification {
|
|
121
|
-
const trimmed = report.trim()
|
|
122
|
-
|
|
123
|
-
// Empty output
|
|
124
|
-
if (NO_OUTPUT_RE.test(trimmed) || trimmed === '') {
|
|
125
|
-
return { status: 'empty' }
|
|
126
|
-
}
|
|
127
|
-
|
|
128
|
-
// Budget-exhausted: extract content before the marker as partialResult
|
|
129
|
-
if (BUDGET_EXHAUSTED_RE.test(trimmed)) {
|
|
130
|
-
const outputSection = extractOutputSection(trimmed)
|
|
131
|
-
const partialResult =
|
|
132
|
-
outputSection !== undefined && !NO_OUTPUT_RE.test(outputSection) ? outputSection : undefined
|
|
133
|
-
|
|
134
|
-
return {
|
|
135
|
-
status: 'budget-exhausted',
|
|
136
|
-
partialResult,
|
|
137
|
-
recommendation:
|
|
138
|
-
'Agent hit step budget; partial result available. ' +
|
|
139
|
-
'Consider resuming with task_id or dispatching with larger budget.',
|
|
140
|
-
}
|
|
141
|
-
}
|
|
142
|
-
|
|
143
|
-
return { status: 'success' }
|
|
144
|
-
}
|
|
145
|
-
|
|
146
|
-
/** Extract the text after "## Output" in a delegation report, or undefined. */
|
|
147
|
-
function extractOutputSection(report: string): string | undefined {
|
|
148
|
-
const marker = '## Output'
|
|
149
|
-
const idx = report.indexOf(marker)
|
|
150
|
-
if (idx === -1) return undefined
|
|
151
|
-
const afterMarker = report.slice(idx + marker.length)
|
|
152
|
-
// Skip the newline after "## Output"
|
|
153
|
-
const content = afterMarker.replace(/^\n+/, '')
|
|
154
|
-
// Stop at the next section (## or **Error**)
|
|
155
|
-
const endIdx = content.search(/\n(?:##|\*\*Error\*\*|\[TIMEOUT)/)
|
|
156
|
-
return endIdx === -1 ? content.trim() : content.slice(0, endIdx).trim()
|
|
157
|
-
}
|
|
158
|
-
|
|
159
|
-
// ─── Timeout Classification ───────────────────────────────────────────
|
|
160
|
-
|
|
161
|
-
/**
|
|
162
|
-
* Classify a timeout based on available output and retry history.
|
|
163
|
-
*
|
|
164
|
-
* - timeout-empty: no report, no messages, not yet retried → retryable
|
|
165
|
-
* - timeout-with-partial: has report or messages → return partial
|
|
166
|
-
* - timeout-exhausted: already retried → escalate
|
|
167
|
-
*
|
|
168
|
-
* @param input Report availability, message presence, and retry count.
|
|
169
|
-
*/
|
|
170
|
-
export function classifyTimeout(input: TimeoutClassifyInput): DelegationResult {
|
|
171
|
-
const { report, hasMessages, retryCount } = input
|
|
172
|
-
const hasPartial =
|
|
173
|
-
(report !== undefined && report.trim() !== '' && !NO_OUTPUT_RE.test(report)) || hasMessages
|
|
174
|
-
|
|
175
|
-
// Already retried → exhausted, no more retries
|
|
176
|
-
if (retryCount >= 1) {
|
|
177
|
-
return {
|
|
178
|
-
status: 'timeout',
|
|
179
|
-
content: report ?? '',
|
|
180
|
-
retryCount,
|
|
181
|
-
subType: 'timeout-exhausted',
|
|
182
|
-
partialResult: hasPartial ? (report ?? '') : undefined,
|
|
183
|
-
recommendation:
|
|
184
|
-
'Agent timed out after retry. Options: ' +
|
|
185
|
-
'(a) try different agent, (b) simplify scope, (c) manual intervention.',
|
|
186
|
-
}
|
|
187
|
-
}
|
|
188
|
-
|
|
189
|
-
// Has partial output → return it
|
|
190
|
-
if (hasPartial) {
|
|
191
|
-
return {
|
|
192
|
-
status: 'timeout',
|
|
193
|
-
content: report ?? '',
|
|
194
|
-
retryCount,
|
|
195
|
-
subType: 'timeout-with-partial',
|
|
196
|
-
partialResult: report ?? '',
|
|
197
|
-
recommendation:
|
|
198
|
-
'Agent timed out but partial work exists. ' + 'Partial result is available above.',
|
|
199
|
-
}
|
|
200
|
-
}
|
|
201
|
-
|
|
202
|
-
// No output at all → retryable
|
|
203
|
-
return {
|
|
204
|
-
status: 'timeout',
|
|
205
|
-
content: report ?? '',
|
|
206
|
-
retryCount,
|
|
207
|
-
subType: 'timeout-empty',
|
|
208
|
-
recommendation: 'Agent timed out with no output. Will retry once with rephrased prompt.',
|
|
209
|
-
}
|
|
210
|
-
}
|
|
211
|
-
|
|
212
|
-
// ─── Result Helpers ───────────────────────────────────────────────────
|
|
213
|
-
|
|
214
|
-
/**
|
|
215
|
-
* Whether a delegation result is retryable (empty + retryCount < 1,
|
|
216
|
-
* or timeout-empty + retryCount < 1).
|
|
217
|
-
*/
|
|
218
|
-
export function isRetryableResult(result: DelegationResult): boolean {
|
|
219
|
-
if (result.retryCount >= 1) return false
|
|
220
|
-
if (result.status === 'empty') return true
|
|
221
|
-
if (result.status === 'timeout' && result.subType === 'timeout-empty') return true
|
|
222
|
-
return false
|
|
223
|
-
}
|
|
224
|
-
|
|
225
|
-
/**
|
|
226
|
-
* Format a DelegationResult into a human-readable string for the orchestrator.
|
|
227
|
-
* Success results are returned as-is (the raw content). Non-success results
|
|
228
|
-
* get a structured status message.
|
|
229
|
-
*/
|
|
230
|
-
export function formatDelegationResult(result: DelegationResult): string {
|
|
231
|
-
if (result.status === 'success') {
|
|
232
|
-
return result.content
|
|
233
|
-
}
|
|
234
|
-
|
|
235
|
-
const parts: string[] = []
|
|
236
|
-
|
|
237
|
-
// Status header
|
|
238
|
-
switch (result.status) {
|
|
239
|
-
case 'empty': {
|
|
240
|
-
const modeLabel =
|
|
241
|
-
result.classification === 'empty-mode2'
|
|
242
|
-
? 'empty-mode2 (reasoning tokens present but no text output)'
|
|
243
|
-
: 'empty-mode1 (no tokens generated)'
|
|
244
|
-
parts.push(`[EMPTY RESULT — ${modeLabel}]`)
|
|
245
|
-
if (result.retryCount === 0) {
|
|
246
|
-
parts.push(
|
|
247
|
-
'The agent returned an empty response. ' +
|
|
248
|
-
'A retry with a rephrased prompt has been triggered automatically.',
|
|
249
|
-
)
|
|
250
|
-
} else {
|
|
251
|
-
parts.push(
|
|
252
|
-
'The agent returned an empty response even after retry. ' +
|
|
253
|
-
'Options: (a) try different agent, (b) simplify scope, (c) manual intervention.',
|
|
254
|
-
)
|
|
255
|
-
}
|
|
256
|
-
break
|
|
257
|
-
}
|
|
258
|
-
case 'budget-exhausted':
|
|
259
|
-
parts.push('[BUDGET EXHAUSTED]')
|
|
260
|
-
if (result.partialResult) {
|
|
261
|
-
parts.push('Partial result:')
|
|
262
|
-
parts.push(result.partialResult)
|
|
263
|
-
}
|
|
264
|
-
parts.push(result.recommendation ?? '')
|
|
265
|
-
break
|
|
266
|
-
case 'timeout': {
|
|
267
|
-
const subLabel =
|
|
268
|
-
result.subType === 'timeout-with-partial'
|
|
269
|
-
? 'with partial output'
|
|
270
|
-
: result.subType === 'timeout-exhausted'
|
|
271
|
-
? 'exhausted after retry'
|
|
272
|
-
: 'no output'
|
|
273
|
-
parts.push(`[TIMEOUT — ${subLabel}]`)
|
|
274
|
-
if (result.partialResult) {
|
|
275
|
-
parts.push('Partial result:')
|
|
276
|
-
parts.push(result.partialResult)
|
|
277
|
-
}
|
|
278
|
-
parts.push(result.recommendation ?? '')
|
|
279
|
-
break
|
|
280
|
-
}
|
|
281
|
-
case 'error':
|
|
282
|
-
parts.push('[ERROR]')
|
|
283
|
-
parts.push(result.content)
|
|
284
|
-
break
|
|
285
|
-
case 'startup_failed':
|
|
286
|
-
parts.push('[STARTUP FAILED]')
|
|
287
|
-
parts.push(result.content)
|
|
288
|
-
parts.push(
|
|
289
|
-
result.recommendation ??
|
|
290
|
-
'Retry once or inspect the child session before dispatching again.',
|
|
291
|
-
)
|
|
292
|
-
break
|
|
293
|
-
case 'bootstrap_unknown':
|
|
294
|
-
parts.push('[BOOTSTRAP UNKNOWN]')
|
|
295
|
-
parts.push(result.content)
|
|
296
|
-
parts.push(
|
|
297
|
-
result.recommendation ?? 'The host did not expose enough session state to verify startup.',
|
|
298
|
-
)
|
|
299
|
-
break
|
|
300
|
-
}
|
|
301
|
-
|
|
302
|
-
// Structured fields for non-success results
|
|
303
|
-
parts.push(`status: ${result.status}`)
|
|
304
|
-
parts.push(`retryCount: ${result.retryCount}`)
|
|
305
|
-
|
|
306
|
-
return parts.filter((p) => p !== '').join('\n')
|
|
307
|
-
}
|