pantheon-opencode 2.0.2 โ 2.0.4
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/AGENTS.global.md +1 -1
- package/AGENTS.md +1 -1
- package/CHANGELOG.md +50 -0
- package/README.md +5 -1
- package/README.pt-BR.md +4 -0
- package/package.json +1 -1
- package/plugin.json +1 -1
- package/pyproject.toml +1 -1
- package/scripts/decision-router-budget.mjs +27 -6
- package/scripts/decision-router.mjs +76 -34
- package/scripts/doctor.mjs +74 -0
- package/scripts/install/decision-router-config.mjs +32 -7
- package/skills-lock.json +1 -1
- package/src/agents/apollo.md +6 -53
- package/src/agents/athena.md +31 -23
- package/src/agents/talos.md +10 -40
- package/src/agents/themis.md +8 -0
- package/src/agents/zeus.md +11 -6
- package/src/pantheon/background-job-board.ts +33 -0
- package/src/pantheon/bus.ts +190 -0
- package/src/pantheon/execution-receipt.ts +113 -0
- package/src/pantheon/jev-decision-tool.ts +8 -2
- package/src/pantheon/local-decision-budget.ts +36 -6
- package/src/pantheon/model-command.ts +2 -4
- package/src/pantheon/prompt-cache-order.ts +89 -0
- package/src/pantheon/themis-reflex.ts +164 -0
- package/src/plugin/index.js +60 -21
- package/src/plugins/tui/dist/tui.js +19 -19
- 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 +1 -1
- package/src/plugins/tui/src/autonomy-state.ts +6 -8
- package/src/plugins/tui/src/decision-router-state.ts +5 -5
- package/src/plugins/tui/src/j1-gate-status.ts +4 -6
- package/src/skills/prompt-refinery/SKILL.md +4 -4
package/src/agents/athena.md
CHANGED
|
@@ -26,6 +26,7 @@ permission:
|
|
|
26
26
|
grep: allow
|
|
27
27
|
glob: allow
|
|
28
28
|
webfetch: allow
|
|
29
|
+
question: allow
|
|
29
30
|
task:
|
|
30
31
|
"*": deny
|
|
31
32
|
apollo: allow
|
|
@@ -110,17 +111,19 @@ Use `webfetch` to verify availability before proposing model updates to other ag
|
|
|
110
111
|
- Make multiple planning attempts
|
|
111
112
|
- Search or inspect `.pantheon/**`, `.pantheon/deepwork/**`, `.git/**` or `.pantheon-worktrees/**`
|
|
112
113
|
|
|
113
|
-
## Plan Structure (
|
|
114
|
+
## Plan Structure (Didactic Executive Summary + Disk Blueprint)
|
|
114
115
|
|
|
115
|
-
|
|
116
|
-
visual, actionable, and token-efficient โ modeled after Claude Code and ChatGPT Agent plan modes:
|
|
116
|
+
Whenever creating a plan, format your output in two clear, modular sections:
|
|
117
117
|
|
|
118
118
|
```markdown
|
|
119
119
|
# ๐ Architecture Plan: [Feature Name]
|
|
120
120
|
|
|
121
121
|
> **Architect:** @athena ยท **Review Gate:** @themis ยท **Execution Mode:** Strict TDD with Sub-10s Test Slicing
|
|
122
122
|
|
|
123
|
-
###
|
|
123
|
+
### ๐ Executive Summary (Chat Contract)
|
|
124
|
+
- **Goal:** One crisp sentence describing the concrete outcome.
|
|
125
|
+
- **Estimated Effort:** [X minutes / low or medium risk]
|
|
126
|
+
|
|
124
127
|
```mermaid
|
|
125
128
|
flowchart LR
|
|
126
129
|
P1["Phase 1: Model / Schema (@demeter)"] --> P2["Phase 2: Core Logic (@hermes)"]
|
|
@@ -128,36 +131,41 @@ flowchart LR
|
|
|
128
131
|
P3 --> P4["Phase 4: Verification (@themis)"]
|
|
129
132
|
```
|
|
130
133
|
|
|
131
|
-
|
|
134
|
+
| Phase | Specialist | Key Deliverable | Test Slice |
|
|
135
|
+
|---|---|---|---|
|
|
136
|
+
| **1. Data / Contract** | @demeter | Schemas, models & migrations | `npm test tests/models/...` |
|
|
137
|
+
| **2. Core Logic** | @hermes | Service implementation & business rules | `npm test tests/services/...` |
|
|
138
|
+
| **3. Interface / UI** | @aphrodite | Components, styling & accessibility | `npm test tests/ui/...` |
|
|
139
|
+
| **4. Verification** | @themis | Security review & full integration gate | `npm run test:e2e` |
|
|
140
|
+
|
|
141
|
+
---
|
|
142
|
+
|
|
143
|
+
### ๐ ๏ธ Technical Implementation Blueprint (Saved to Disk)
|
|
144
|
+
|
|
145
|
+
#### Goal, Scope & Context
|
|
132
146
|
- **Goal:** One sentence describing what this plan achieves.
|
|
133
|
-
- **In scope:**
|
|
134
|
-
- **
|
|
135
|
-
- **Assumptions:** [Key assumptions; unverified facts]
|
|
136
|
-
- **Open Questions:** [Critical decisions the user must make, with recommended defaults]
|
|
147
|
+
- **In scope / Out of scope:** Concrete outcomes and exclusions.
|
|
148
|
+
- **Assumptions & Open Questions:** Key assumptions; critical decisions with recommended defaults.
|
|
137
149
|
|
|
138
|
-
|
|
150
|
+
#### Phases & Quality Gates (3-5 max)
|
|
139
151
|
1. **[Phase Name] โ @agent**
|
|
140
152
|
- **Goal & Actions:** Outcome and concrete steps (TDD: RED โ GREEN โ REFACTOR)
|
|
141
153
|
- **Affected files or areas:** Known paths or explicitly mark unknown paths instead of guessing
|
|
142
154
|
- **Dependencies:** Earlier phases or external systems
|
|
143
|
-
- **Tests & Test Slicing:** Exact targeted slice (
|
|
144
|
-
- **Risks & Mitigation:** Specific risk โ concrete
|
|
145
|
-
|
|
146
|
-
2. **[Phase Name] โ @agent**
|
|
147
|
-
...
|
|
155
|
+
- **Tests & Test Slicing:** Exact targeted slice (< 10s), focused checks, lint-first gate
|
|
156
|
+
- **Risks & Mitigation:** Specific risk โ concrete prevention
|
|
148
157
|
|
|
149
|
-
|
|
158
|
+
#### Acceptance Criteria & Verification
|
|
150
159
|
- [ ] Observable completion criterion
|
|
151
|
-
- [ ] Verification criterion (
|
|
160
|
+
- [ ] Verification criterion (targeted tests green, 0 lint warnings)
|
|
152
161
|
|
|
153
|
-
|
|
154
|
-
- **Failure Trigger:** Conditions under which execution aborts
|
|
155
|
-
- **Rollback Steps:** Revert commit / feature flag toggle / database rollback migration
|
|
162
|
+
#### Rollback & Contingency Plan
|
|
163
|
+
- **Failure Trigger:** Conditions under which execution aborts
|
|
164
|
+
- **Rollback Steps:** Revert commit / feature flag toggle / database rollback migration
|
|
156
165
|
```
|
|
157
166
|
|
|
158
|
-
|
|
159
|
-
|
|
160
|
-
"approval requested", or content available only in your child-session transcript.
|
|
167
|
+
### Interactive Question Gate
|
|
168
|
+
When directly invoked by the user, if you need trade-off confirmation or final approval, always call the native `question` tool with structured choices (e.g., `["Aprovar e executar", "Ajustar plano"]`), rather than relying solely on unstructured chat text.
|
|
161
169
|
Keep unknown paths and unverified assumptions explicit rather than fabricating
|
|
162
170
|
detail.
|
|
163
171
|
|
package/src/agents/talos.md
CHANGED
|
@@ -51,32 +51,11 @@ mcp_tools:
|
|
|
51
51
|
- **Code-mode / infrastructure reads** โ cheap and on demand; never for
|
|
52
52
|
decisions the ladder already answers.
|
|
53
53
|
|
|
54
|
-
##
|
|
54
|
+
## Fast-Path Execution Contract
|
|
55
55
|
|
|
56
|
-
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
## Core Capabilities
|
|
60
|
-
|
|
61
|
-
### 1. Rapid Repairs
|
|
62
|
-
- Single-file fixes (< 10 lines)
|
|
63
|
-
- Multi-file fixes (max 2 files)
|
|
64
|
-
- CSS, typo, import, and minor logic fixes
|
|
65
|
-
|
|
66
|
-
### 2. No TDD Ceremony
|
|
67
|
-
- Hotfixes skip the RED->GREEN->REFACTOR cycle
|
|
68
|
-
- Fix and run the narrowest relevant existing test/check; do not run a full suite for a micro-edit
|
|
69
|
-
- Document the root cause inline
|
|
70
|
-
|
|
71
|
-
### 3. Escalation Rules
|
|
72
|
-
Escalate to @zeus if:
|
|
73
|
-
- Fix requires > 2 files or > 10 lines changed
|
|
74
|
-
- Has security implications
|
|
75
|
-
- Requires database migration
|
|
76
|
-
- Touches authentication, security-sensitive behavior, production data, global configuration, or broad permissions
|
|
77
|
-
- Breaks existing tests unexpectedly
|
|
78
|
-
|
|
79
|
-
Full-auto does not permit Talos to perform push/merge/deploy, destructive/global changes, or sensitive work; escalate for required specialist review and explicit human approval.
|
|
56
|
+
- **Micro-scope only**: single-file (< 10 lines) or max 2 files (CSS, typos, imports, isolated logic fixes).
|
|
57
|
+
- **Zero ceremony**: skip RED->GREEN->REFACTOR; run only the narrowest relevant test slice via `test-slicing`.
|
|
58
|
+
- **Standalone execution**: no subagents, no orchestration overhead. Return `## subtask_summary` directly.
|
|
80
59
|
|
|
81
60
|
## Requested Artifact Completion
|
|
82
61
|
|
|
@@ -84,19 +63,10 @@ Full-auto does not permit Talos to perform push/merge/deploy, destructive/global
|
|
|
84
63
|
- Searching the MCP Code Mode catalog only discovers MCP tools. A missing MCP match does not mean native file tools are unavailable.
|
|
85
64
|
- Verify the write succeeded and, when practical, read the artifact back. If a native file tool is actually unavailable or denies the write, state that specific limitation and do not claim completion.
|
|
86
65
|
|
|
87
|
-
##
|
|
88
|
-
- No orchestration: you work standalone
|
|
89
|
-
- No Themis review needed (low-risk)
|
|
90
|
-
- Return subtask_summary format
|
|
91
|
-
- If complexity exceeds threshold, escalate immediately
|
|
92
|
-
|
|
93
|
-
## Auto-Continue (Embedded: Hotfix)
|
|
94
|
-
|
|
95
|
-
- Auto-continue through quick fix cycles (identify โ fix โ verify)
|
|
96
|
-
- No checkpoint needed (single-file fixes, low complexity)
|
|
97
|
-
- Escalate to Zeus if fix takes > 3 turns or requires > 2 files / > 10 lines
|
|
98
|
-
- If fix breaks existing tests, stop immediately and escalate
|
|
99
|
-
- No partial results โ either fix is applied or escalate
|
|
66
|
+
## Escalation Boundaries
|
|
100
67
|
|
|
101
|
-
|
|
102
|
-
|
|
68
|
+
Escalate to @zeus immediately if:
|
|
69
|
+
- Change requires > 2 files or > 10 lines
|
|
70
|
+
- Touches security, authentication, credentials, migrations, or broad permissions
|
|
71
|
+
- Fix takes > 3 turns or causes unexpected test regression
|
|
72
|
+
- Full-auto never permits push/merge/deploy or destructive operations without specialist review and human approval.
|
package/src/agents/themis.md
CHANGED
|
@@ -183,6 +183,14 @@ Estimated time: ~60% of review effort
|
|
|
183
183
|
- [ ] IntentGate passed (code solves the right problem)
|
|
184
184
|
- [ ] Auth/security, data/schema, and sensitive actions preserve applicable tests, specialist review, and human approval
|
|
185
185
|
|
|
186
|
+
## Local Reflex Loop (Self-Healing)
|
|
187
|
+
|
|
188
|
+
When Layer 1 (lint/formatting/style) or Layer 2 (narrow test slice) fails:
|
|
189
|
+
- Themis executes a direct local reflex loop with the implementer (up to 2 retry attempts).
|
|
190
|
+
- Emits `themis:reflex:discrepancy` with the exact failing slice/rules.
|
|
191
|
+
- Does NOT escalate to Zeus unless the 2-retry reflex budget is exhausted or an architectural/IntentGate mismatch occurs.
|
|
192
|
+
- On successful remediation, emits `themis:reflex:resolved` and issues PASS / PASS_WITH_NOTES.
|
|
193
|
+
|
|
186
194
|
## Verdict
|
|
187
195
|
|
|
188
196
|
After review, return one of these:
|
package/src/agents/zeus.md
CHANGED
|
@@ -11,6 +11,7 @@ permission:
|
|
|
11
11
|
vision_describe: allow
|
|
12
12
|
pantheon_jev_decide: allow
|
|
13
13
|
pantheon_skill_search: allow
|
|
14
|
+
question: allow
|
|
14
15
|
task:
|
|
15
16
|
"*": allow
|
|
16
17
|
temperature: 0.2
|
|
@@ -97,9 +98,13 @@ Before every dispatch, send the target a concise brief with the objective/result
|
|
|
97
98
|
## Plan Handoff and Approval
|
|
98
99
|
|
|
99
100
|
Whenever the user explicitly asks for a plan, or starts `/pantheon-deepwork`:
|
|
100
|
-
1. Delegate planning to @athena synchronously via `subagent`. Tell Athena to return the
|
|
101
|
-
2. Check that the returned subagent result contains
|
|
102
|
-
3.
|
|
103
|
-
4.
|
|
104
|
-
|
|
105
|
-
|
|
101
|
+
1. Delegate planning to @athena synchronously via `subagent`. Tell Athena to return both the Executive Summary and the full technical plan in the subagent result; Zeus owns the user-facing approval gate.
|
|
102
|
+
2. Check that the returned subagent result contains a valid plan with goal, scope, actionable phases, affected files, verification, risks, and acceptance criteria.
|
|
103
|
+
3. Write the full technical blueprint to `.pantheon/deepwork/<slug>/PLAN.md` immediately so the complete specification is permanently available on disk for implementation specialists.
|
|
104
|
+
4. Present a **didactic, visual Executive Decision Summary in chat (1 screen max, ~15-20 lines)**:
|
|
105
|
+
- **Goal & Target:** 1 crisp sentence.
|
|
106
|
+
- **Execution Flow:** Mermaid diagram (3-4 nodes).
|
|
107
|
+
- **Slice Table:** Compact table (`Phase | Specialist | Key Deliverable | Verification Slice`).
|
|
108
|
+
- **Disk Blueprint:** Clear reference that the exhaustive technical spec is preserved in `.pantheon/deepwork/<slug>/PLAN.md`.
|
|
109
|
+
5. **Interactive Question Gate:** Always invoke the native `question` tool to request user approval with explicit options (e.g., `["Aprovar e iniciar execuรงรฃo", "Ajustar plano"]` / `["Approve and begin execution", "Adjust plan"]`), falling back to plain text only if the tool is not exposed.
|
|
110
|
+
6. If the user requests changes, send feedback to Athena, update `.pantheon/deepwork/<slug>/PLAN.md`, and repeat the interactive gate.
|
|
@@ -13,6 +13,7 @@
|
|
|
13
13
|
import { mkdir, rename, unlink, writeFile } from 'node:fs/promises'
|
|
14
14
|
import { dirname, join } from 'node:path'
|
|
15
15
|
|
|
16
|
+
import { getSharedBus } from './bus.ts'
|
|
16
17
|
import { createPantheonLogger } from './logger.ts'
|
|
17
18
|
|
|
18
19
|
// Silence-by-default TUI policy (pantheon-hooks L42-58): console output in a
|
|
@@ -292,6 +293,14 @@ export class BackgroundJobBoard {
|
|
|
292
293
|
|
|
293
294
|
this.jobs.set(input.taskID, record)
|
|
294
295
|
await this.persistRecord(record)
|
|
296
|
+
getSharedBus().emit('job:created', {
|
|
297
|
+
taskID: record.taskID,
|
|
298
|
+
alias: record.alias,
|
|
299
|
+
agent: record.agent,
|
|
300
|
+
parentSessionID: record.parentSessionID,
|
|
301
|
+
state: record.state,
|
|
302
|
+
timestamp: now,
|
|
303
|
+
})
|
|
295
304
|
return record
|
|
296
305
|
}
|
|
297
306
|
|
|
@@ -364,6 +373,14 @@ export class BackgroundJobBoard {
|
|
|
364
373
|
|
|
365
374
|
// --- write-ahead log: persist before notify ---
|
|
366
375
|
await this.persistRecord(job)
|
|
376
|
+
getSharedBus().emit('job:updated', {
|
|
377
|
+
taskID: job.taskID,
|
|
378
|
+
alias: job.alias,
|
|
379
|
+
agent: job.agent,
|
|
380
|
+
state: job.state,
|
|
381
|
+
resultSummary: job.resultSummary,
|
|
382
|
+
timestamp: job.updatedAt,
|
|
383
|
+
})
|
|
367
384
|
|
|
368
385
|
// --- signal on terminal states ---
|
|
369
386
|
if (willBeTerminal) {
|
|
@@ -463,6 +480,11 @@ export class BackgroundJobBoard {
|
|
|
463
480
|
// cycle deleted them. Instead, remove any still-pending wake signal so
|
|
464
481
|
// reconcile leaves no `<alias>.signal.json` behind.
|
|
465
482
|
await this.deleteSignal(job.alias)
|
|
483
|
+
getSharedBus().emit('job:reconciled', {
|
|
484
|
+
taskID: job.taskID,
|
|
485
|
+
alias: job.alias,
|
|
486
|
+
timestamp: job.updatedAt,
|
|
487
|
+
})
|
|
466
488
|
|
|
467
489
|
return job
|
|
468
490
|
}
|
|
@@ -728,6 +750,17 @@ export class BackgroundJobBoard {
|
|
|
728
750
|
}
|
|
729
751
|
|
|
730
752
|
private notifyTerminal(taskID: string): void {
|
|
753
|
+
const job = this.jobs.get(taskID)
|
|
754
|
+
if (job) {
|
|
755
|
+
getSharedBus().emit('job:terminal', {
|
|
756
|
+
taskID: job.taskID,
|
|
757
|
+
alias: job.alias,
|
|
758
|
+
agent: job.agent,
|
|
759
|
+
state: job.state,
|
|
760
|
+
resultSummary: job.resultSummary,
|
|
761
|
+
timestamp: job.completedAt ?? Date.now(),
|
|
762
|
+
})
|
|
763
|
+
}
|
|
731
764
|
for (const listener of this.terminalListeners) {
|
|
732
765
|
try {
|
|
733
766
|
listener(taskID)
|
|
@@ -0,0 +1,190 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Pantheon Event Bus (pantheon-bus) โ In-memory reactive event system.
|
|
3
|
+
*
|
|
4
|
+
* Eliminates disk I/O and polling loops for inter-agent communication,
|
|
5
|
+
* background job tracking, and reactive verification.
|
|
6
|
+
*
|
|
7
|
+
* Process-global singleton anchored in globalThis.__pantheonSharedBus,
|
|
8
|
+
* matching the dual-plugin deduplication architecture in shared-board.ts.
|
|
9
|
+
*
|
|
10
|
+
* Pure TypeScript โ zero external dependencies.
|
|
11
|
+
*
|
|
12
|
+
* @module bus
|
|
13
|
+
*/
|
|
14
|
+
|
|
15
|
+
export interface JobEventPayload {
|
|
16
|
+
readonly taskID: string
|
|
17
|
+
readonly alias: string
|
|
18
|
+
readonly agent: string
|
|
19
|
+
readonly timestamp: number
|
|
20
|
+
readonly parentSessionID?: string | undefined
|
|
21
|
+
readonly state?: string | undefined
|
|
22
|
+
readonly resultSummary?: string | null | undefined
|
|
23
|
+
}
|
|
24
|
+
|
|
25
|
+
export interface ThemisReflexEventPayload {
|
|
26
|
+
readonly taskID: string
|
|
27
|
+
readonly agent: string
|
|
28
|
+
readonly file: string
|
|
29
|
+
readonly error: string
|
|
30
|
+
readonly attempt: number
|
|
31
|
+
readonly maxAttempts: number
|
|
32
|
+
readonly timestamp: number
|
|
33
|
+
readonly resolved?: boolean | undefined
|
|
34
|
+
}
|
|
35
|
+
|
|
36
|
+
export interface ExecutionReceiptPayload {
|
|
37
|
+
readonly receiptId: string
|
|
38
|
+
readonly taskID: string
|
|
39
|
+
readonly agent: string
|
|
40
|
+
readonly status: 'VERIFIED' | 'UNVERIFIED' | 'FAILED'
|
|
41
|
+
readonly files: ReadonlyArray<{
|
|
42
|
+
readonly path: string
|
|
43
|
+
readonly hash: string
|
|
44
|
+
readonly sizeBytes: number
|
|
45
|
+
}>
|
|
46
|
+
readonly testSlice?:
|
|
47
|
+
| { readonly command: string; readonly exitCode: number; readonly durationMs: number }
|
|
48
|
+
| undefined
|
|
49
|
+
readonly timestamp: number
|
|
50
|
+
readonly signature: string
|
|
51
|
+
}
|
|
52
|
+
|
|
53
|
+
export interface PantheonEventMap {
|
|
54
|
+
'job:created': JobEventPayload
|
|
55
|
+
'job:updated': JobEventPayload
|
|
56
|
+
'job:terminal': JobEventPayload
|
|
57
|
+
'job:reconciled': { readonly taskID: string; readonly alias: string; readonly timestamp: number }
|
|
58
|
+
'themis:reflex:discrepancy': ThemisReflexEventPayload
|
|
59
|
+
'themis:reflex:resolved': ThemisReflexEventPayload
|
|
60
|
+
'agent:execution:receipt': ExecutionReceiptPayload
|
|
61
|
+
}
|
|
62
|
+
|
|
63
|
+
export type PantheonEventName = keyof PantheonEventMap
|
|
64
|
+
|
|
65
|
+
// biome-ignore lint/suspicious/noExplicitAny: internal heterogeneous event listener storage
|
|
66
|
+
type AnyListener = (payload: any) => void | Promise<void>
|
|
67
|
+
|
|
68
|
+
export class PantheonBus {
|
|
69
|
+
private readonly listeners: Map<string, Set<AnyListener>> = new Map()
|
|
70
|
+
|
|
71
|
+
/**
|
|
72
|
+
* Register a persistent listener for an event.
|
|
73
|
+
* Returns an unsubscribe callback for easy cleanup.
|
|
74
|
+
*/
|
|
75
|
+
on<E extends PantheonEventName>(
|
|
76
|
+
event: E,
|
|
77
|
+
listener: (payload: PantheonEventMap[E]) => void | Promise<void>,
|
|
78
|
+
): () => void {
|
|
79
|
+
let set = this.listeners.get(event)
|
|
80
|
+
if (!set) {
|
|
81
|
+
set = new Set()
|
|
82
|
+
this.listeners.set(event, set)
|
|
83
|
+
}
|
|
84
|
+
set.add(listener)
|
|
85
|
+
return () => {
|
|
86
|
+
set?.delete(listener)
|
|
87
|
+
if (set?.size === 0) {
|
|
88
|
+
this.listeners.delete(event)
|
|
89
|
+
}
|
|
90
|
+
}
|
|
91
|
+
}
|
|
92
|
+
|
|
93
|
+
/**
|
|
94
|
+
* Register a one-shot listener that unregisters itself after firing once.
|
|
95
|
+
*/
|
|
96
|
+
once<E extends PantheonEventName>(
|
|
97
|
+
event: E,
|
|
98
|
+
listener: (payload: PantheonEventMap[E]) => void | Promise<void>,
|
|
99
|
+
): () => void {
|
|
100
|
+
const unsubscribe = this.on(event, (payload) => {
|
|
101
|
+
unsubscribe()
|
|
102
|
+
return listener(payload)
|
|
103
|
+
})
|
|
104
|
+
return unsubscribe
|
|
105
|
+
}
|
|
106
|
+
|
|
107
|
+
/**
|
|
108
|
+
* Synchronously emit an event to all registered listeners.
|
|
109
|
+
* Listener rejections are logged safely without throwing.
|
|
110
|
+
*/
|
|
111
|
+
emit<E extends PantheonEventName>(event: E, payload: PantheonEventMap[E]): void {
|
|
112
|
+
const set = this.listeners.get(event)
|
|
113
|
+
if (!set || set.size === 0) return
|
|
114
|
+
for (const listener of Array.from(set)) {
|
|
115
|
+
try {
|
|
116
|
+
const result = listener(payload)
|
|
117
|
+
if (result instanceof Promise) {
|
|
118
|
+
result.catch(() => {})
|
|
119
|
+
}
|
|
120
|
+
} catch {
|
|
121
|
+
// Safe swallow: an in-memory listener crash must not fail the emitter
|
|
122
|
+
}
|
|
123
|
+
}
|
|
124
|
+
}
|
|
125
|
+
|
|
126
|
+
/**
|
|
127
|
+
* Wait asynchronously for a specific event matching a predicate, with an optional timeout.
|
|
128
|
+
*/
|
|
129
|
+
waitFor<E extends PantheonEventName>(
|
|
130
|
+
event: E,
|
|
131
|
+
predicate: (payload: PantheonEventMap[E]) => boolean,
|
|
132
|
+
timeoutMs = 30_000,
|
|
133
|
+
): Promise<PantheonEventMap[E]> {
|
|
134
|
+
return new Promise((resolve, reject) => {
|
|
135
|
+
let timer: NodeJS.Timeout | null = null
|
|
136
|
+
let unsubscribe: (() => void) | null = null
|
|
137
|
+
|
|
138
|
+
if (timeoutMs > 0 && Number.isFinite(timeoutMs)) {
|
|
139
|
+
timer = setTimeout(() => {
|
|
140
|
+
unsubscribe?.()
|
|
141
|
+
reject(new Error(`Timed out waiting for bus event "${event}" after ${timeoutMs}ms`))
|
|
142
|
+
}, timeoutMs)
|
|
143
|
+
}
|
|
144
|
+
|
|
145
|
+
unsubscribe = this.on(event, (payload) => {
|
|
146
|
+
try {
|
|
147
|
+
if (predicate(payload)) {
|
|
148
|
+
if (timer) clearTimeout(timer)
|
|
149
|
+
unsubscribe?.()
|
|
150
|
+
resolve(payload)
|
|
151
|
+
}
|
|
152
|
+
} catch (err) {
|
|
153
|
+
if (timer) clearTimeout(timer)
|
|
154
|
+
unsubscribe?.()
|
|
155
|
+
reject(err)
|
|
156
|
+
}
|
|
157
|
+
})
|
|
158
|
+
})
|
|
159
|
+
}
|
|
160
|
+
|
|
161
|
+
/** Number of active listeners for a specific event or in total. */
|
|
162
|
+
listenerCount(event?: PantheonEventName): number {
|
|
163
|
+
if (event) {
|
|
164
|
+
return this.listeners.get(event)?.size ?? 0
|
|
165
|
+
}
|
|
166
|
+
let total = 0
|
|
167
|
+
for (const set of this.listeners.values()) {
|
|
168
|
+
total += set.size
|
|
169
|
+
}
|
|
170
|
+
return total
|
|
171
|
+
}
|
|
172
|
+
|
|
173
|
+
/** Clear all listeners (primarily for unit tests). */
|
|
174
|
+
clear(): void {
|
|
175
|
+
this.listeners.clear()
|
|
176
|
+
}
|
|
177
|
+
}
|
|
178
|
+
|
|
179
|
+
const BUS_KEY = '__pantheonSharedBus'
|
|
180
|
+
|
|
181
|
+
/**
|
|
182
|
+
* Access the process-global PantheonBus singleton.
|
|
183
|
+
*/
|
|
184
|
+
export function getSharedBus(): PantheonBus {
|
|
185
|
+
const host = globalThis as { [BUS_KEY]?: PantheonBus }
|
|
186
|
+
if (host[BUS_KEY] instanceof PantheonBus) return host[BUS_KEY]
|
|
187
|
+
const bus = new PantheonBus()
|
|
188
|
+
host[BUS_KEY] = bus
|
|
189
|
+
return bus
|
|
190
|
+
}
|
|
@@ -0,0 +1,113 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Execution Proof Receipts โ Verifiable task completion contracts for agent execution.
|
|
3
|
+
*
|
|
4
|
+
* Implements SOTA late-2026 verification principles: subagents return cryptographic
|
|
5
|
+
* receipts containing deterministic content hashes of modified files and verifiable
|
|
6
|
+
* test-slicing proofs, eliminating false-positive completion reports.
|
|
7
|
+
*
|
|
8
|
+
* Emits reactive verification events on `PantheonBus`.
|
|
9
|
+
*
|
|
10
|
+
* Pure TypeScript โ stdlib crypto only.
|
|
11
|
+
*
|
|
12
|
+
* @module execution-receipt
|
|
13
|
+
*/
|
|
14
|
+
|
|
15
|
+
import { createHash } from 'node:crypto'
|
|
16
|
+
import { type ExecutionReceiptPayload, getSharedBus } from './bus.ts'
|
|
17
|
+
|
|
18
|
+
export interface FileInput {
|
|
19
|
+
readonly path: string
|
|
20
|
+
readonly content: string | Buffer
|
|
21
|
+
}
|
|
22
|
+
|
|
23
|
+
export interface TestSliceInput {
|
|
24
|
+
readonly command: string
|
|
25
|
+
readonly exitCode: number
|
|
26
|
+
readonly durationMs: number
|
|
27
|
+
}
|
|
28
|
+
|
|
29
|
+
export interface CreateReceiptInput {
|
|
30
|
+
readonly taskID: string
|
|
31
|
+
readonly agent: string
|
|
32
|
+
readonly files: readonly FileInput[]
|
|
33
|
+
readonly testSlice?: TestSliceInput | undefined
|
|
34
|
+
}
|
|
35
|
+
|
|
36
|
+
/**
|
|
37
|
+
* Compute SHA-256 fingerprint for a string or buffer.
|
|
38
|
+
*/
|
|
39
|
+
export function hashContent(content: string | Buffer): string {
|
|
40
|
+
return createHash('sha256').update(content).digest('hex')
|
|
41
|
+
}
|
|
42
|
+
|
|
43
|
+
/**
|
|
44
|
+
* Compute cryptographic signature binding task, agent, timestamp and file hashes.
|
|
45
|
+
*/
|
|
46
|
+
export function computeReceiptSignature(
|
|
47
|
+
taskID: string,
|
|
48
|
+
agent: string,
|
|
49
|
+
timestamp: number,
|
|
50
|
+
files: ReadonlyArray<{ readonly path: string; readonly hash: string }>,
|
|
51
|
+
): string {
|
|
52
|
+
const sortedFiles = [...files].sort((a, b) => a.path.localeCompare(b.path))
|
|
53
|
+
const payload = [
|
|
54
|
+
taskID,
|
|
55
|
+
agent,
|
|
56
|
+
String(timestamp),
|
|
57
|
+
sortedFiles.map((f) => `${f.path}:${f.hash}`).join(';'),
|
|
58
|
+
].join('|')
|
|
59
|
+
return createHash('sha256').update(payload).digest('hex')
|
|
60
|
+
}
|
|
61
|
+
|
|
62
|
+
/**
|
|
63
|
+
* Generate a verifiable Execution Proof Receipt and emit it on the shared bus.
|
|
64
|
+
*/
|
|
65
|
+
export function createExecutionReceipt(input: CreateReceiptInput): ExecutionReceiptPayload {
|
|
66
|
+
const timestamp = Date.now()
|
|
67
|
+
const files = input.files.map((f) => {
|
|
68
|
+
const buf = typeof f.content === 'string' ? Buffer.from(f.content, 'utf8') : f.content
|
|
69
|
+
return {
|
|
70
|
+
path: f.path,
|
|
71
|
+
hash: hashContent(buf),
|
|
72
|
+
sizeBytes: buf.length,
|
|
73
|
+
}
|
|
74
|
+
})
|
|
75
|
+
|
|
76
|
+
const signature = computeReceiptSignature(input.taskID, input.agent, timestamp, files)
|
|
77
|
+
const receiptId = `rcpt_${signature.slice(0, 16)}`
|
|
78
|
+
|
|
79
|
+
let status: 'VERIFIED' | 'UNVERIFIED' | 'FAILED' = 'UNVERIFIED'
|
|
80
|
+
if (input.testSlice) {
|
|
81
|
+
status = input.testSlice.exitCode === 0 ? 'VERIFIED' : 'FAILED'
|
|
82
|
+
}
|
|
83
|
+
|
|
84
|
+
const receipt: ExecutionReceiptPayload = {
|
|
85
|
+
receiptId,
|
|
86
|
+
taskID: input.taskID,
|
|
87
|
+
agent: input.agent,
|
|
88
|
+
status,
|
|
89
|
+
files,
|
|
90
|
+
testSlice: input.testSlice,
|
|
91
|
+
timestamp,
|
|
92
|
+
signature,
|
|
93
|
+
}
|
|
94
|
+
|
|
95
|
+
getSharedBus().emit('agent:execution:receipt', receipt)
|
|
96
|
+
return receipt
|
|
97
|
+
}
|
|
98
|
+
|
|
99
|
+
/**
|
|
100
|
+
* Verify whether an execution receipt signature matches its declared files and metadata.
|
|
101
|
+
*/
|
|
102
|
+
export function verifyExecutionReceipt(receipt: ExecutionReceiptPayload): boolean {
|
|
103
|
+
if (!receipt?.signature || !receipt.taskID || !receipt.agent) {
|
|
104
|
+
return false
|
|
105
|
+
}
|
|
106
|
+
const expected = computeReceiptSignature(
|
|
107
|
+
receipt.taskID,
|
|
108
|
+
receipt.agent,
|
|
109
|
+
receipt.timestamp,
|
|
110
|
+
receipt.files,
|
|
111
|
+
)
|
|
112
|
+
return receipt.signature === expected
|
|
113
|
+
}
|
|
@@ -288,8 +288,14 @@ export async function runGatedNativeJevCall(params: {
|
|
|
288
288
|
let config: ReturnType<typeof readUserDecisionRouterConfig>
|
|
289
289
|
try {
|
|
290
290
|
config = readUserDecisionRouterConfig()
|
|
291
|
-
} catch {
|
|
292
|
-
|
|
291
|
+
} catch (error) {
|
|
292
|
+
const detail =
|
|
293
|
+
error instanceof Error && error.message.includes('group-writable')
|
|
294
|
+
? 'CONFIG_PATH_GROUP_WRITABLE'
|
|
295
|
+
: error instanceof Error
|
|
296
|
+
? error.name
|
|
297
|
+
: 'ConfigError'
|
|
298
|
+
return { ok: false, reasonCode: 'CONFIG_UNAVAILABLE', detail }
|
|
293
299
|
}
|
|
294
300
|
if (!config.enabled) return { ok: false, reasonCode: 'DISABLED' }
|
|
295
301
|
if (!config.egress_consent) return { ok: false, reasonCode: 'CONSENT_MISSING' }
|
|
@@ -1,5 +1,13 @@
|
|
|
1
1
|
import { randomUUID } from 'node:crypto'
|
|
2
|
-
import {
|
|
2
|
+
import {
|
|
3
|
+
chmodSync,
|
|
4
|
+
closeSync,
|
|
5
|
+
existsSync,
|
|
6
|
+
lstatSync,
|
|
7
|
+
mkdirSync,
|
|
8
|
+
openSync,
|
|
9
|
+
type Stats,
|
|
10
|
+
} from 'node:fs'
|
|
3
11
|
import { homedir } from 'node:os'
|
|
4
12
|
import { basename, dirname, isAbsolute, join, parse, resolve, sep } from 'node:path'
|
|
5
13
|
import { performance } from 'node:perf_hooks'
|
|
@@ -1202,20 +1210,42 @@ function ensureTrustedDirectoryChain(directoryPath: string): void {
|
|
|
1202
1210
|
}
|
|
1203
1211
|
}
|
|
1204
1212
|
|
|
1213
|
+
function isPermittedGroupWritable(env: NodeJS.ProcessEnv = process.env): boolean {
|
|
1214
|
+
if (env.PANTHEON_ALLOW_GROUP_WRITABLE === '1' || env.PANTHEON_ALLOW_GROUP_WRITABLE === 'true') {
|
|
1215
|
+
return true
|
|
1216
|
+
}
|
|
1217
|
+
if (
|
|
1218
|
+
env.JUPYTERHUB_USER ||
|
|
1219
|
+
env.JUPYTER_IMAGE ||
|
|
1220
|
+
env.JUPYTER_SERVER_ROOT ||
|
|
1221
|
+
env.KUBERNETES_SERVICE_HOST ||
|
|
1222
|
+
env.container ||
|
|
1223
|
+
existsSync('/.dockerenv')
|
|
1224
|
+
) {
|
|
1225
|
+
return true
|
|
1226
|
+
}
|
|
1227
|
+
return false
|
|
1228
|
+
}
|
|
1229
|
+
|
|
1205
1230
|
function validateDirectoryTrust(stat: Stats, uid: number, isFinalDirectory: boolean): void {
|
|
1206
1231
|
if (!stat.isDirectory() || stat.isSymbolicLink()) {
|
|
1207
1232
|
throw new Error('Budget ledger path contains a non-directory or symlink ancestor')
|
|
1208
1233
|
}
|
|
1209
|
-
|
|
1234
|
+
const mode = stat.mode & 0o7777
|
|
1235
|
+
const isSticky = (mode & 0o1000) !== 0
|
|
1236
|
+
const isSystemOrUser = stat.uid === uid || stat.uid === 0 || (isSticky && stat.uid === 65534)
|
|
1237
|
+
if (!isSystemOrUser) {
|
|
1210
1238
|
throw new Error('Budget ledger path contains an untrusted owner')
|
|
1211
1239
|
}
|
|
1212
1240
|
if (isFinalDirectory && stat.uid !== uid) {
|
|
1213
1241
|
throw new Error('Budget ledger directory must be owned by the current user')
|
|
1214
1242
|
}
|
|
1215
|
-
const
|
|
1216
|
-
const
|
|
1217
|
-
|
|
1218
|
-
|
|
1243
|
+
const worldWritable = (mode & 0o002) !== 0
|
|
1244
|
+
const groupWritable = (mode & 0o020) !== 0
|
|
1245
|
+
if (worldWritable && !isSticky) {
|
|
1246
|
+
throw new Error('Budget ledger path contains a group/other-writable directory')
|
|
1247
|
+
}
|
|
1248
|
+
if (groupWritable && !isSticky && !isFinalDirectory && !isPermittedGroupWritable()) {
|
|
1219
1249
|
throw new Error('Budget ledger path contains a group/other-writable directory')
|
|
1220
1250
|
}
|
|
1221
1251
|
}
|
|
@@ -606,7 +606,7 @@ async function updateActivePreset(
|
|
|
606
606
|
if (!options.v2Mode) {
|
|
607
607
|
const env = envValue(options)
|
|
608
608
|
fallbackPreset = defaultFallbackPreset()
|
|
609
|
-
// Prefer env preset if set and known
|
|
609
|
+
// Prefer env preset if set and known.
|
|
610
610
|
const envPreset = env.PANTHEON_MODEL_PRESET?.trim()
|
|
611
611
|
if (envPreset && envPreset !== '' && envPreset !== 'none') {
|
|
612
612
|
try {
|
|
@@ -1031,9 +1031,7 @@ export function createModelCommand(options: ModelCommandOptions = {}): ModelTool
|
|
|
1031
1031
|
const args = options.v2Mode
|
|
1032
1032
|
? {
|
|
1033
1033
|
...modelArgs,
|
|
1034
|
-
action: modelArgs.action.describe(
|
|
1035
|
-
'Operation: status/show, set, or reset. V1 no-args invocation runs the legacy wizard.',
|
|
1036
|
-
),
|
|
1034
|
+
action: modelArgs.action.describe('Operation: status/show, set, or reset.'),
|
|
1037
1035
|
}
|
|
1038
1036
|
: modelArgs
|
|
1039
1037
|
|