pantheon-opencode 1.5.0-beta.2 → 1.5.0-beta.20
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/.pantheon/code-mode/manifest.json +21 -0
- package/AGENTS.md +14 -63
- package/CHANGELOG.md +1260 -0
- package/README.md +162 -23
- package/README.pt-BR.md +104 -19
- package/bin/pantheon-init.mjs +178 -21
- package/commands/pantheon.md +1 -35
- package/package.json +23 -15
- package/plugin.json +1 -1
- package/pyproject.toml +62 -0
- package/scripts/code_mode_server.py +431 -55
- package/scripts/doctor.mjs +207 -9
- package/scripts/install/cli-ui.mjs +17 -11
- package/scripts/install/config-migration.mjs +4 -6
- package/scripts/install/opencode.mjs +137 -35
- package/scripts/install/plugin.mjs +12 -45
- package/scripts/install/shared.mjs +43 -2
- package/scripts/install/strings.mjs +129 -0
- package/scripts/install-mcp.mjs +1 -0
- package/scripts/manifest-inventory.mjs +410 -0
- package/scripts/mcp_persistence_server.py +536 -164
- package/scripts/package-evidence.mjs +242 -0
- package/scripts/probe-pantheon-cost.mjs +10 -1
- package/scripts/release-beta-version.mjs +56 -60
- package/scripts/sync-artifacts.mjs +169 -0
- package/scripts/sync-tui.mjs +129 -64
- package/scripts/test-opencode-v1-v2-sandbox.sh +79 -161
- package/scripts/validation-policy.mjs +7 -7
- package/scripts/version-check.mjs +113 -108
- package/scripts/versioning.mjs +230 -86
- package/src/agents/apollo.md +1 -0
- package/src/agents/gaia.md +1 -0
- package/src/agents/mnemosyne.md +1 -1
- package/src/agents/zeus.md +11 -54
- package/src/instructions/memory-protocol.instructions.md +6 -3
- package/src/instructions/zeus-anti-stall.instructions.md +2 -4
- package/src/instructions/zeus-council-synthesis.instructions.md +3 -28
- package/src/instructions/zeus-timeout-retry.instructions.md +3 -28
- package/src/mcp/code_mode_server.py +431 -55
- package/src/mcp/mcp_persistence_server.py +536 -164
- package/src/mcp/toon_codec.py +2 -4
- package/src/pantheon/background-job-board.ts +47 -5
- package/src/pantheon/{delegation-compaction.ts → compaction-context.ts} +10 -100
- package/src/pantheon/cost-command.ts +309 -44
- package/src/pantheon/delegation-enforce.ts +59 -14
- package/src/pantheon/file-persistence-salvage.ts +64 -0
- package/src/pantheon/file-persistence.ts +194 -45
- package/src/pantheon/goal-loop.ts +1 -1
- package/src/pantheon/goal-store.ts +1 -1
- package/src/pantheon/idle-continuation.ts +4 -6
- package/src/pantheon/model-command.ts +1 -2
- package/src/pantheon/native-task-status.ts +139 -0
- package/src/pantheon/presets.d.mts +0 -20
- package/src/pantheon/presets.mjs +0 -64
- package/src/pantheon/session-guard.ts +2 -2
- package/src/pantheon/session-id.ts +30 -0
- package/src/pantheon/shared-board.ts +37 -0
- package/src/pantheon/task-result-guard.ts +106 -0
- package/src/pantheon/todo-enforcer.ts +3 -3
- package/src/pantheon/tool-ceiling.ts +159 -0
- package/src/pantheon/tool-context.ts +23 -0
- package/src/pantheon/v2-bridge.ts +1 -4
- package/src/plugin-v2.ts +8 -21
- package/src/plugin.ts +83 -251
- package/src/plugins/pantheon-hooks.ts +51 -0
- package/src/plugins/tui/dist/server.d.ts +1 -1
- package/src/plugins/tui/dist/server.d.ts.map +1 -1
- package/src/plugins/tui/dist/server.js +1 -1
- package/src/plugins/tui/dist/server.js.map +1 -1
- package/src/plugins/tui/dist/tui.d.ts +261 -77
- package/src/plugins/tui/dist/tui.d.ts.map +1 -1
- package/src/plugins/tui/dist/tui.js +640 -234
- package/src/plugins/tui/dist/tui.js.map +1 -1
- package/src/plugins/tui/package-lock.json +2 -2
- package/src/plugins/tui/package.json +2 -2
- package/src/plugins/tui/src/index.tsx +872 -226
- package/src/routing.yml +0 -16
- package/src/skills/context-compression/SKILL.md +1 -1
- package/scripts/generate-preset-docs.mjs +0 -300
- package/scripts/generate-prompts.sh +0 -222
- package/scripts/hash_verify.py +0 -192
- package/scripts/install/component.mjs +0 -24
- package/scripts/install/migrate-config.mjs +0 -302
- package/scripts/test-context-compression.sh +0 -169
- package/src/pantheon/auto-wake.ts +0 -89
- package/src/pantheon/chat-reminders.ts +0 -9
- package/src/pantheon/comment-checker.ts +0 -48
- package/src/pantheon/compaction-assert.ts +0 -160
- package/src/pantheon/cost-tracker.ts +0 -108
- package/src/pantheon/delegate-manager.ts +0 -669
- package/src/pantheon/delegation-classifier.ts +0 -307
- package/src/pantheon/delegation-finalize.ts +0 -374
- package/src/pantheon/delegation-notify.ts +0 -212
- package/src/pantheon/delegation-toast.ts +0 -73
- package/src/pantheon/delegation.ts +0 -1120
- package/src/pantheon/dispatch-guard.ts +0 -109
- package/src/pantheon/permission-eval.ts +0 -284
- package/src/pantheon/preemptive-compact.ts +0 -76
- package/src/pantheon/retry-policy.ts +0 -212
- package/src/pantheon/step-cap.ts +0 -110
- package/src/pantheon/token-opt.ts +0 -781
- package/src/pantheon/toon-codec.ts +0 -481
- package/src/pantheon/v2-events.ts +0 -131
- package/src/pantheon/v2-hooks.ts +0 -287
- package/src/pantheon/v2-tool-definitions.ts +0 -331
- package/src/pantheon/zeus-delegate-with-retry.ts +0 -406
- package/src/plugins/tui/dist/tui.tsx +0 -2605
|
@@ -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
|
-
}
|
|
@@ -1,374 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Delegation finalize — PULL output from a child session and persist the
|
|
3
|
-
* delegation report (markdown) + the terminal board transition.
|
|
4
|
-
*
|
|
5
|
-
* Completion is NOT pushed: the event hook observes `session.idle` on the
|
|
6
|
-
* child session, then calls `finalizeDelegation()` which pulls the child's
|
|
7
|
-
* messages via `client.session.messages()`, writes the report atomically
|
|
8
|
-
* (tmp + rename) to `.pantheon/delegations/<rootSessionID>/<alias>.md`, and
|
|
9
|
-
* moves the job to a terminal board state (persist-before-notify: the report
|
|
10
|
-
* file exists before `board.updateStatus` resolves waiters).
|
|
11
|
-
*
|
|
12
|
-
* Pure TypeScript — zero runtime dependencies beyond Node.js builtins and the
|
|
13
|
-
* BackgroundJobBoard. The client is a STRUCTURAL interface (fake-client
|
|
14
|
-
* friendly) so this module is testable standalone, without the opencode SDK.
|
|
15
|
-
*
|
|
16
|
-
* @module delegation-finalize
|
|
17
|
-
*/
|
|
18
|
-
|
|
19
|
-
import { mkdir, readFile, rename, writeFile } from 'node:fs/promises'
|
|
20
|
-
import { join } from 'node:path'
|
|
21
|
-
|
|
22
|
-
import type { BackgroundJobBoard, BackgroundJobRecord } from './background-job-board.ts'
|
|
23
|
-
import type { PermissionTaskConfig } from './permission-globs.ts'
|
|
24
|
-
import type { StepCapTracker } from './step-cap.ts'
|
|
25
|
-
|
|
26
|
-
// ─── Types ─────────────────────────────────────────────────────────────
|
|
27
|
-
|
|
28
|
-
/** A minimal structural view of one message bundle from session.messages(). */
|
|
29
|
-
export interface DelegationMessageBundle {
|
|
30
|
-
info?: { role?: string; finish?: string; error?: unknown }
|
|
31
|
-
parts?: Array<{
|
|
32
|
-
type?: string
|
|
33
|
-
text?: string
|
|
34
|
-
/** Tool name on `tool` parts — used by the activity sampler. */
|
|
35
|
-
tool?: string
|
|
36
|
-
/** Tool input on `tool` parts (opencode SDK ToolPart.metadata). */
|
|
37
|
-
metadata?: { input?: unknown }
|
|
38
|
-
}>
|
|
39
|
-
}
|
|
40
|
-
|
|
41
|
-
/** Minimal structural view of the opencode SDK session client we use. */
|
|
42
|
-
export interface DelegationClientSession {
|
|
43
|
-
create(input: {
|
|
44
|
-
body: { parentID: string; title?: string; model?: { id: string; providerID: string } }
|
|
45
|
-
}): Promise<{ id: string }>
|
|
46
|
-
promptAsync(input: {
|
|
47
|
-
path: { id: string }
|
|
48
|
-
body: {
|
|
49
|
-
agent: string
|
|
50
|
-
model?: { id: string; providerID: string }
|
|
51
|
-
parts: Array<{ type: 'text'; text: string }>
|
|
52
|
-
}
|
|
53
|
-
/** Request cancellation for hosts that support AbortSignal. */
|
|
54
|
-
signal?: AbortSignal
|
|
55
|
-
}): Promise<unknown>
|
|
56
|
-
messages?(input: { path: { id: string } }): Promise<Array<DelegationMessageBundle>>
|
|
57
|
-
/** Optional session metadata endpoint (older hosts do not expose it). */
|
|
58
|
-
get?(input: { path: { id: string } }): Promise<unknown>
|
|
59
|
-
}
|
|
60
|
-
|
|
61
|
-
/** The client surface used by the delegation core. */
|
|
62
|
-
export interface DelegationClient {
|
|
63
|
-
session: DelegationClientSession
|
|
64
|
-
}
|
|
65
|
-
|
|
66
|
-
/** Tunables for the delegation core (all optional — see DELEGATION_DEFAULTS). */
|
|
67
|
-
export interface DelegationOptions {
|
|
68
|
-
/** Job timeout before the delegation is finalized as error/timedOut. */
|
|
69
|
-
timeoutMs?: number
|
|
70
|
-
/** Caller-side cap for pantheon_delegation_read's waitForTerminal. */
|
|
71
|
-
readTimeoutMs?: number
|
|
72
|
-
/** Base directory for delegation reports (default `.pantheon/delegations`). */
|
|
73
|
-
outputDir?: string
|
|
74
|
-
/** Explicit root-session allowlist for the depth guard. */
|
|
75
|
-
rootSessions?: ReadonlySet<string>
|
|
76
|
-
/** Custom depth-guard predicate — overrides rootSessions and the default. */
|
|
77
|
-
isRootSession?: (sessionID: string) => boolean
|
|
78
|
-
/** SDK-backed child predicate; authoritative when supplied by the plugin. */
|
|
79
|
-
isChildSession?: (sessionID: string) => boolean
|
|
80
|
-
/**
|
|
81
|
-
* Enforce the runtime B2 delegation matrix; createDelegationTools defaults
|
|
82
|
-
* this to true. Set false only for an explicitly legacy host that cannot
|
|
83
|
-
* provide ToolContext.agent.
|
|
84
|
-
*/
|
|
85
|
-
enforceRuntimeMatrix?: boolean
|
|
86
|
-
/** Record a child returned by the SDK session.create call. */
|
|
87
|
-
registerChildSession?: (sessionID: string, parentID: string) => void
|
|
88
|
-
/**
|
|
89
|
-
* Agents whose delegated child sessions are registered as read-only
|
|
90
|
-
* (Phase 4 enforcement). Mirrors routing.yml
|
|
91
|
-
* background_delegation.read_only_agents. Matching is case-insensitive.
|
|
92
|
-
*/
|
|
93
|
-
readOnlyAgents?: ReadonlySet<string>
|
|
94
|
-
/**
|
|
95
|
-
* Per-agent model overrides for delegated child sessions, keyed by agent
|
|
96
|
-
* name (lowercase, e.g. `apollo`). Values are opencode model IDs in
|
|
97
|
-
* `provider/model` form (e.g. `opencode/deepseek-v4-flash-free`), sourced
|
|
98
|
-
* from routing.yml agent entries. Higher priority than the active preset.
|
|
99
|
-
*/
|
|
100
|
-
agentModels?: Readonly<Record<string, string>>
|
|
101
|
-
/**
|
|
102
|
-
* Fase B3: Per-exception delegation budgets. Keys use
|
|
103
|
-
* `<parent-agent>-><target-agent>` (for example `athena->apollo`). The
|
|
104
|
-
* count is tracked per parent session for this process. It is intentionally
|
|
105
|
-
* not persisted across restart because this layer has no reliable store.
|
|
106
|
-
* An absent key has no budget limit.
|
|
107
|
-
*/
|
|
108
|
-
delegationBudgets?: ReadonlyMap<string, number>
|
|
109
|
-
/**
|
|
110
|
-
* Fase B3: Wall-clock timeout (ms) for individual delegation sessions.
|
|
111
|
-
* The plugin resolves `PANTHEON_DELEGATION_TIMEOUT_MS` into this option;
|
|
112
|
-
* direct callers may provide it here instead.
|
|
113
|
-
*/
|
|
114
|
-
wallClockTimeoutMs?: number
|
|
115
|
-
/**
|
|
116
|
-
* Env override for the active-preset resolution (see resolveActivePreset).
|
|
117
|
-
* Defaults to process.env. Primarily for tests.
|
|
118
|
-
*/
|
|
119
|
-
presetEnv?: Record<string, string | undefined>
|
|
120
|
-
/** Warning sink for the no-model fallback (defaults to console). */
|
|
121
|
-
logger?: { warn?: (msg: string) => void }
|
|
122
|
-
/**
|
|
123
|
-
* R4: per-agent step-cap tracker (routing.yml `agents.<name>.max_steps`).
|
|
124
|
-
* When an agent reaches max_steps the delegation is forced to
|
|
125
|
-
* summarize-and-stop: skipped with a capped summary when already at the
|
|
126
|
-
* cap, or the prompt is appended with a stop instruction when the cap is
|
|
127
|
-
* hit by this dispatch.
|
|
128
|
-
*/
|
|
129
|
-
stepCapTracker?: StepCapTracker
|
|
130
|
-
/**
|
|
131
|
-
* O5: permission.task glob rules controlling which subagents may be
|
|
132
|
-
* invoked. Denied agents are removed from the delegate tool description
|
|
133
|
-
* entirely (not just blocked at call time).
|
|
134
|
-
*/
|
|
135
|
-
permissionTask?: PermissionTaskConfig
|
|
136
|
-
/**
|
|
137
|
-
* O5: candidate agent names for the delegate tool description. Defaults
|
|
138
|
-
* to the canonical 14-agent roster.
|
|
139
|
-
*/
|
|
140
|
-
agentNames?: readonly string[]
|
|
141
|
-
/** Bounded post-prompt bootstrap watchdog window. */
|
|
142
|
-
bootstrapTimeoutMs?: number
|
|
143
|
-
/** Timeout for the promptAsync request itself (including AbortSignal). */
|
|
144
|
-
promptTimeoutMs?: number
|
|
145
|
-
/** Poll interval used by the bootstrap watchdog. */
|
|
146
|
-
bootstrapPollIntervalMs?: number
|
|
147
|
-
/** Injectable clock/sleeper for deterministic watchdog tests. */
|
|
148
|
-
now?: () => number
|
|
149
|
-
sleep?: (ms: number) => Promise<void>
|
|
150
|
-
}
|
|
151
|
-
|
|
152
|
-
/** Dependencies threaded to the finalize path. */
|
|
153
|
-
export interface DelegationDeps {
|
|
154
|
-
board: BackgroundJobBoard
|
|
155
|
-
client: DelegationClient
|
|
156
|
-
options: DelegationOptions
|
|
157
|
-
}
|
|
158
|
-
|
|
159
|
-
/** Terminal transition requested by the finalize path. */
|
|
160
|
-
export interface FinalizeInput {
|
|
161
|
-
state: 'completed' | 'error' | 'startup_failed' | 'startup_unknown' | 'cancelled'
|
|
162
|
-
error?: string
|
|
163
|
-
timedOut?: boolean
|
|
164
|
-
}
|
|
165
|
-
|
|
166
|
-
/** Defaults matching routing.yml background_delegation (timeout_ms 900000). */
|
|
167
|
-
export const DELEGATION_DEFAULTS = {
|
|
168
|
-
timeoutMs: 900_000,
|
|
169
|
-
readTimeoutMs: 900_000,
|
|
170
|
-
outputDir: '.pantheon/delegations',
|
|
171
|
-
} as const
|
|
172
|
-
|
|
173
|
-
// ─── Output pulling ────────────────────────────────────────────────────
|
|
174
|
-
|
|
175
|
-
interface PulledOutput {
|
|
176
|
-
text: string
|
|
177
|
-
/** True only when the child produced an assistant message or tool activity. */
|
|
178
|
-
hasWork: boolean
|
|
179
|
-
}
|
|
180
|
-
|
|
181
|
-
/** Pull visible text and structurally identify real child work. */
|
|
182
|
-
async function pullOutput(client: DelegationClient, childSessionID: string): Promise<PulledOutput> {
|
|
183
|
-
if (typeof client.session.messages !== 'function')
|
|
184
|
-
return { text: '(child session messages API unavailable)', hasWork: false }
|
|
185
|
-
try {
|
|
186
|
-
const bundles = await client.session.messages({ path: { id: childSessionID } })
|
|
187
|
-
const lines: string[] = []
|
|
188
|
-
let hasWork = false
|
|
189
|
-
for (const bundle of bundles) {
|
|
190
|
-
const hasToolPart = (bundle.parts ?? []).some((part) => part.type === 'tool')
|
|
191
|
-
const hasTextPart = (bundle.parts ?? []).some(
|
|
192
|
-
(part) => part.type === 'text' && typeof part.text === 'string' && part.text.trim() !== '',
|
|
193
|
-
)
|
|
194
|
-
if (
|
|
195
|
-
bundle.info?.role === 'assistant' &&
|
|
196
|
-
bundle.info.error === undefined &&
|
|
197
|
-
(hasToolPart || hasTextPart)
|
|
198
|
-
) {
|
|
199
|
-
hasWork = true
|
|
200
|
-
}
|
|
201
|
-
for (const part of bundle.parts ?? []) {
|
|
202
|
-
if (
|
|
203
|
-
part.type === 'tool' &&
|
|
204
|
-
bundle.info?.role === 'assistant' &&
|
|
205
|
-
bundle.info.error === undefined
|
|
206
|
-
)
|
|
207
|
-
hasWork = true
|
|
208
|
-
if (part.type === 'text' && typeof part.text === 'string' && part.text.trim() !== '') {
|
|
209
|
-
lines.push(part.text)
|
|
210
|
-
}
|
|
211
|
-
}
|
|
212
|
-
}
|
|
213
|
-
return { text: lines.join('\n\n'), hasWork }
|
|
214
|
-
} catch (err: unknown) {
|
|
215
|
-
const reason = err instanceof Error ? err.message : String(err)
|
|
216
|
-
return { text: `(failed to pull child session output: ${reason})`, hasWork: false }
|
|
217
|
-
}
|
|
218
|
-
}
|
|
219
|
-
|
|
220
|
-
// ─── Markdown rendering ────────────────────────────────────────────────
|
|
221
|
-
|
|
222
|
-
/** Render the delegation report body. Public for tests/Phase 3 reuse. */
|
|
223
|
-
export function renderDelegationMarkdown(
|
|
224
|
-
job: BackgroundJobRecord,
|
|
225
|
-
output: string,
|
|
226
|
-
opts: FinalizeInput & { at?: number },
|
|
227
|
-
): string {
|
|
228
|
-
const at = opts.at ?? Date.now()
|
|
229
|
-
const lines = [
|
|
230
|
-
`# Delegation Report — ${job.alias}`,
|
|
231
|
-
'',
|
|
232
|
-
`- **Task ID**: \`${job.taskID}\``,
|
|
233
|
-
`- **Agent**: ${job.agent}`,
|
|
234
|
-
`- **Description**: ${job.description}`,
|
|
235
|
-
`- **State**: ${opts.state}`,
|
|
236
|
-
`- **Timed out**: ${opts.timedOut === true ? 'true' : 'false'}`,
|
|
237
|
-
`- **Started**: ${new Date(job.launchedAt).toISOString()}`,
|
|
238
|
-
`- **Finalized**: ${new Date(at).toISOString()}`,
|
|
239
|
-
'',
|
|
240
|
-
'## Output',
|
|
241
|
-
'',
|
|
242
|
-
output.trim() !== '' ? output : '_No output captured._',
|
|
243
|
-
]
|
|
244
|
-
if (opts.error !== undefined) {
|
|
245
|
-
lines.push('', `**Error**: ${opts.error}`)
|
|
246
|
-
}
|
|
247
|
-
if (opts.timedOut === true) {
|
|
248
|
-
lines.push('', '[TIMEOUT REACHED]')
|
|
249
|
-
}
|
|
250
|
-
return `${lines.join('\n')}\n`
|
|
251
|
-
}
|
|
252
|
-
|
|
253
|
-
/** One-line result summary from the pulled output (for the board record). */
|
|
254
|
-
function summarizeOutput(output: string): string {
|
|
255
|
-
const firstLine = output
|
|
256
|
-
.split('\n')
|
|
257
|
-
.map((l) => l.trim())
|
|
258
|
-
.find((l) => l !== '')
|
|
259
|
-
if (firstLine === undefined) return 'No output captured'
|
|
260
|
-
return firstLine.length > 120 ? `${firstLine.slice(0, 120)}…` : firstLine
|
|
261
|
-
}
|
|
262
|
-
|
|
263
|
-
// ─── Report persistence ────────────────────────────────────────────────
|
|
264
|
-
|
|
265
|
-
/**
|
|
266
|
-
* Reject session IDs that could escape the delegation output directory.
|
|
267
|
-
* `parentSessionID` is embedded in the report path
|
|
268
|
-
* `<outputDir>/<parentSessionID>/<alias>.md` — `path.join` does NOT strip
|
|
269
|
-
* `..`, so a traversing ID would write outside the sandbox. Called BEFORE any
|
|
270
|
-
* path construction (F1 — path traversal hardening).
|
|
271
|
-
*
|
|
272
|
-
* @throws {Error} When the ID contains `..`, `/`, or `\`.
|
|
273
|
-
*/
|
|
274
|
-
export function assertSafeParentSessionID(parentSessionID: string): void {
|
|
275
|
-
if (
|
|
276
|
-
parentSessionID.includes('..') ||
|
|
277
|
-
parentSessionID.includes('/') ||
|
|
278
|
-
parentSessionID.includes('\\')
|
|
279
|
-
) {
|
|
280
|
-
throw new Error(`Invalid parentSessionID: ${parentSessionID}`)
|
|
281
|
-
}
|
|
282
|
-
}
|
|
283
|
-
|
|
284
|
-
/** Atomic write (tmp + rename) of the report under <outputDir>/<parent>/<alias>.md. */
|
|
285
|
-
async function writeDelegationReport(
|
|
286
|
-
outputDir: string,
|
|
287
|
-
job: BackgroundJobRecord,
|
|
288
|
-
md: string,
|
|
289
|
-
): Promise<string> {
|
|
290
|
-
assertSafeParentSessionID(job.parentSessionID)
|
|
291
|
-
const dir = join(outputDir, job.parentSessionID)
|
|
292
|
-
const filePath = join(dir, `${job.alias}.md`)
|
|
293
|
-
const tmpPath = join(dir, `${job.alias}.md.tmp`)
|
|
294
|
-
await mkdir(dir, { recursive: true })
|
|
295
|
-
await writeFile(tmpPath, md, 'utf-8')
|
|
296
|
-
await rename(tmpPath, filePath)
|
|
297
|
-
return filePath
|
|
298
|
-
}
|
|
299
|
-
|
|
300
|
-
/** Read a previously written report; undefined when the file does not exist. */
|
|
301
|
-
export async function readDelegationReport(
|
|
302
|
-
outputDir: string,
|
|
303
|
-
job: BackgroundJobRecord,
|
|
304
|
-
): Promise<string | undefined> {
|
|
305
|
-
assertSafeParentSessionID(job.parentSessionID)
|
|
306
|
-
const filePath = join(outputDir, job.parentSessionID, `${job.alias}.md`)
|
|
307
|
-
try {
|
|
308
|
-
return await readFile(filePath, 'utf-8')
|
|
309
|
-
} catch (err: unknown) {
|
|
310
|
-
if (err instanceof Error && 'code' in err && (err as NodeJS.ErrnoException).code === 'ENOENT') {
|
|
311
|
-
return undefined
|
|
312
|
-
}
|
|
313
|
-
throw err
|
|
314
|
-
}
|
|
315
|
-
}
|
|
316
|
-
|
|
317
|
-
// ─── Finalize ──────────────────────────────────────────────────────────
|
|
318
|
-
|
|
319
|
-
/**
|
|
320
|
-
* Finalize a delegation for a child session: pull output, write the report
|
|
321
|
-
* atomically, then move the board job to the requested terminal state.
|
|
322
|
-
*
|
|
323
|
-
* The board transition is guarded: only `running` → terminal, or an
|
|
324
|
-
* idempotent same-terminal re-notify, are applied. A finalize that races a
|
|
325
|
-
* timeout (already `error`) never overwrites the terminal state — the report
|
|
326
|
-
* file is still (re)written, so the marker reflects the latest observation.
|
|
327
|
-
*
|
|
328
|
-
* @returns The updated job record, or undefined when the session is unknown.
|
|
329
|
-
*/
|
|
330
|
-
export async function finalizeDelegation(
|
|
331
|
-
deps: DelegationDeps,
|
|
332
|
-
childSessionID: string,
|
|
333
|
-
opts: FinalizeInput,
|
|
334
|
-
): Promise<BackgroundJobRecord | undefined> {
|
|
335
|
-
const job = deps.board.get(childSessionID)
|
|
336
|
-
if (!job) return undefined
|
|
337
|
-
|
|
338
|
-
const pulled = await pullOutput(deps.client, childSessionID)
|
|
339
|
-
const output = pulled.text
|
|
340
|
-
const effectiveOpts: FinalizeInput =
|
|
341
|
-
opts.state === 'completed' && !pulled.hasWork
|
|
342
|
-
? {
|
|
343
|
-
state: 'error',
|
|
344
|
-
error: opts.error ?? 'Child session produced no assistant or tool output',
|
|
345
|
-
...(opts.timedOut !== undefined ? { timedOut: opts.timedOut } : {}),
|
|
346
|
-
}
|
|
347
|
-
: opts
|
|
348
|
-
const outputDir = deps.options.outputDir ?? DELEGATION_DEFAULTS.outputDir
|
|
349
|
-
await writeDelegationReport(outputDir, job, renderDelegationMarkdown(job, output, effectiveOpts))
|
|
350
|
-
|
|
351
|
-
const status: {
|
|
352
|
-
taskID: string
|
|
353
|
-
state: 'completed' | 'error' | 'startup_failed' | 'startup_unknown' | 'cancelled'
|
|
354
|
-
error?: string
|
|
355
|
-
timedOut?: boolean
|
|
356
|
-
resultSummary?: string
|
|
357
|
-
} = {
|
|
358
|
-
taskID: childSessionID,
|
|
359
|
-
state: effectiveOpts.state,
|
|
360
|
-
resultSummary: summarizeOutput(output),
|
|
361
|
-
}
|
|
362
|
-
if (effectiveOpts.error !== undefined) status.error = effectiveOpts.error
|
|
363
|
-
if (effectiveOpts.timedOut !== undefined) status.timedOut = effectiveOpts.timedOut
|
|
364
|
-
|
|
365
|
-
const current = deps.board.get(childSessionID)
|
|
366
|
-
if (current !== undefined) {
|
|
367
|
-
if (current.state === 'running' || current.state === effectiveOpts.state) {
|
|
368
|
-
await deps.board.updateStatus(status)
|
|
369
|
-
}
|
|
370
|
-
// Terminal in a DIFFERENT state (e.g. timeout already recorded) — the
|
|
371
|
-
// report above still captures this finalize; the board state stands.
|
|
372
|
-
}
|
|
373
|
-
return deps.board.get(childSessionID) ?? job
|
|
374
|
-
}
|