agent-crdt-synchronizer
Синхронизация данных между узлами через CRDT
Что получите
Реализация CRDT-синхронизации с правилами слияния изменений и учётом причинного порядка операций.
Что передать
Описание узлов, структуры данных, начальное состояние и требования к согласованности.
Проверен состав файлов. Запуск не тестировался. Что проверено
Оценки пользователей
Как использовать
Помогает разработчикам распределённых приложений реализовать слияние параллельных изменений через CRDT — структуры данных с правилами разрешения конфликтов. Рассматривает обмен полным состоянием, отдельными операциями и накопленными изменениями.
- Укажите, какие данные синхронизируются: счётчики, наборы элементов, отдельные значения или последовательности.
- Опишите допустимые операции и ожидаемое поведение при одновременных изменениях на разных узлах.
- Уточните, достаточно ли постепенного совпадения копий или для части операций требуется согласование между узлами.
Условия работы
- Основной сценарий предполагает постепенное совпадение копий данных; строгая согласованность рассматривается отдельно, через интеграцию с протоколом консенсуса.
- Некоторые примеры намеренно упрощены: порядок и слияние последовательностей, а также вычисление изменений требуют доработки для конкретной системы.
Пример запроса после установки
Помоги спроектировать синхронизацию списка задач между несколькими узлами: пользователи одновременно добавляют и удаляют элементы, а после обмена изменениями копии должны совпасть.
Пример составлен по инструкции автора; результат запуска не проверялся.
Основание: инструкция закреплённой версииДля установки
Node.js, npm и Git для команды выше. Подключения и инструменты для работы скилла настраиваются отдельно.
Совместимые агенты
Совместимость указана для инструкций. Запуск в каждом приложении не тестировался; для веб-чата способ подключения может отличаться.
Файлы скилла
GitHubЧитать инструкцию SKILL.md
Исходный текст закреплённой версии.
---
name: agent-crdt-synchronizer
description: Agent skill for crdt-synchronizer - invoke with $agent-crdt-synchronizer
---
---
name: crdt-synchronizer
type: synchronizer
color: "#4CAF50"
description: Implements Conflict-free Replicated Data Types for eventually consistent state synchronization
capabilities:
- state_based_crdts
- operation_based_crdts
- delta_synchronization
- conflict_resolution
- causal_consistency
priority: high
hooks:
pre: |
echo "🔄 CRDT Synchronizer syncing: $TASK"
# Initialize CRDT state tracking
if [[ "$TASK" == *"synchronization"* ]]; then
echo "📊 Preparing delta state computation"
fi
post: |
echo "🎯 CRDT synchronization complete"
# Verify eventual consistency
echo "✅ Validating conflict-free state convergence"
---
# CRDT Synchronizer
Implements Conflict-free Replicated Data Types for eventually consistent distributed state synchronization.
## Core Responsibilities
1. **CRDT Implementation**: Deploy state-based and operation-based conflict-free data types
2. **Data Structure Management**: Handle counters, sets, registers, and composite structures
3. **Delta Synchronization**: Implement efficient incremental state updates
4. **Conflict Resolution**: Ensure deterministic conflict-free merge operations
5. **Causal Consistency**: Maintain proper ordering of causally related operations
## Technical Implementation
### Base CRDT Framework
```javascript
class CRDTSynchronizer {
constructor(nodeId, replicationGroup) {
this.nodeId = nodeId;
this.replicationGroup = replicationGroup;
this.crdtInstances = new Map();
this.vectorClock = new VectorClock(nodeId);
this.deltaBuffer = new Map();
this.syncScheduler = new SyncScheduler();
this.causalTracker = new CausalTracker();
}
// Register CRDT instance
registerCRDT(name, crdtType, initialState = null) {
const crdt = this.createCRDTInstance(crdtType, initialState);
this.crdtInstances.set(name, crdt);
// Subscribe to CRDT changes for delta tracking
crdt.onUpdate((delta) => {
this.trackDelta(name, delta);
});
return crdt;
}
// Create specific CRDT instance
createCRDTInstance(type, initialState) {
switch (type) {
case 'G_COUNTER':
return new GCounter(this.nodeId, this.replicationGroup, initialState);
case 'PN_COUNTER':
return new PNCounter(this.nodeId, this.replicationGroup, initialState);
case 'OR_SET':
return new ORSet(this.nodeId, initialState);
case 'LWW_REGISTER':
return new LWWRegister(this.nodeId, initialState);
case 'OR_MAP':
return new ORMap(this.nodeId, this.replicationGroup, initialState);
case 'RGA':
return new RGA(this.nodeId, initialState);
default:
throw new Error(`Unknown CRDT type: ${type}`);
}
}
// Synchronize with peer nodes
async synchronize(peerNodes = null) {
const targets = peerNodes || Array.from(this.replicationGroup);
for (const peer of targets) {
if (peer !== this.nodeId) {
await this.synchronizeWithPeer(peer);
}
}
}
async synchronizeWithPeer(peerNode) {
// Get current state and deltas
const localState = this.getCurrentState();
const deltas = this.getDeltasSince(peerNode);
// Send sync request
const syncRequest = {
type: 'CRDT_SYNC_REQUEST',
sender: this.nodeId,
vectorClock: this.vectorClock.clone(),
state: localState,
deltas: deltas
};
try {
const response = await this.sendSyncRequest(peerNode, syncRequest);
await this.processSyncResponse(response);
} catch (error) {
console.error(`Sync failed with ${peerNode}:`, error);
}
}
}
```
### G-Counter Implementation
```javascript
class GCounter {
constructor(nodeId, replicationGroup, initialState = null) {
this.nodeId = nodeId;
this.replicationGroup = replicationGroup;
this.payload = new Map();
// Initialize counters for all nodes
for (const node of replicationGroup) {
this.payload.set(node, 0);
}
if (initialState) {
this.merge(initialState);
}
this.updateCallbacks = [];
}
// Increment operation (can only be performed by owner node)
increment(amount = 1) {
if (amount < 0) {
throw new Error('G-Counter only supports positive increments');
}
const oldValue = this.payload.get(this.nodeId) || 0;
const newValue = oldValue + amount;
this.payload.set(this.nodeId, newValue);
// Notify observers
this.notifyUpdate({
type: 'INCREMENT',
node: this.nodeId,
oldValue: oldValue,
newValue: newValue,
delta: amount
});
return newValue;
}
// Get current value (sum of all node counters)
value() {
return Array.from(this.payload.values()).reduce((sum, val) => sum + val, 0);
}
// Merge with another G-Counter state
merge(otherState) {
let changed = false;
for (const [node, otherValue] of otherState.payload) {
const currentValue = this.payload.get(node) || 0;
if (otherValue > currentValue) {
this.payload.set(node, otherValue);
changed = true;
}
}
if (changed) {
this.notifyUpdate({
type: 'MERGE',
mergedFrom: otherState
});
}
}
// Compare with another state
compare(otherState) {
for (const [node, otherValue] of otherState.payload) {
const currentValue = this.payload.get(node) || 0;
if (currentValue < otherValue) {
return 'LESS_THAN';
} else if (currentValue > otherValue) {
return 'GREATER_THAN';
}
}
return 'EQUAL';
}
// Clone current state
clone() {
const newCounter = new GCounter(this.nodeId, this.replicationGroup);
newCounter.payload = new Map(this.payload);
return newCounter;
}
onUpdate(callback) {
this.updateCallbacks.push(callback);
}
notifyUpdate(delta) {
this.updateCallbacks.forEach(callback => callback(delta));
}
}
```
### OR-Set Implementation
```javascript
class ORSet {
constructor(nodeId, initialState = null) {
this.nodeId = nodeId;
this.elements = new Map(); // element -> Set of unique tags
this.tombstones = new Set(); // removed element tags
this.tagCounter = 0;
if (initialState) {
this.merge(initialState);
}
this.updateCallbacks = [];
}
// Add element to set
add(element) {
const tag = this.generateUniqueTag();
if (!this.elements.has(element)) {
this.elements.set(element, new Set());
}
this.elements.get(element).add(tag);
this.notifyUpdate({
type: 'ADD',
element: element,
tag: tag
});
return tag;
}
// Remove element from set
remove(element) {
if (!this.elements.has(element)) {
return false; // Element not present
}
const tags = this.elements.get(element);
const removedTags = [];
// Add all tags to tombstones
for (const tag of tags) {
this.tombstones.add(tag);
removedTags.push(tag);
}
this.notifyUpdate({
type: 'REMOVE',
element: element,
removedTags: removedTags
});
return true;
}
// Check if element is in set
has(element) {
if (!this.elements.has(element)) {
return false;
}
const tags = this.elements.get(element);
// Element is present if it has at least one non-tombstoned tag
for (const tag of tags) {
if (!this.tombstones.has(tag)) {
return true;
}
}
return false;
}
// Get all elements in set
values() {
const result = new Set();
for (const [element, tags] of this.elements) {
// Include element if it has at least one non-tombstoned tag
for (const tag of tags) {
if (!this.tombstones.has(tag)) {
result.add(element);
break;
}
}
}
return result;
}
// Merge with another OR-Set
merge(otherState) {
let changed = false;
// Merge elements and their tags
for (const [element, otherTags] of otherState.elements) {
if (!this.elements.has(element)) {
this.elements.set(element, new Set());
}
const currentTags = this.elements.get(element);
for (const tag of otherTags) {
if (!currentTags.has(tag)) {
currentTags.add(tag);
changed = true;
}
}
}
// Merge tombstones
for (const tombstone of otherState.tombstones) {
if (!this.tombstones.has(tombstone)) {
this.tombstones.add(tombstone);
changed = true;
}
}
if (changed) {
this.notifyUpdate({
type: 'MERGE',
mergedFrom: otherState
});
}
}
generateUniqueTag() {
return `${this.nodeId}-${Date.now()}-${++this.tagCounter}`;
}
onUpdate(callback) {
this.updateCallbacks.push(callback);
}
notifyUpdate(delta) {
this.updateCallbacks.forEach(callback => callback(delta));
}
}
```
### LWW-Register Implementation
```javascript
class LWWRegister {
constructor(nodeId, initialValue = null) {
this.nodeId = nodeId;
this.value = initialValue;
this.timestamp = initialValue ? Date.now() : 0;
this.vectorClock = new VectorClock(nodeId);
this.updateCallbacks = [];
}
// Set new value with timestamp
set(newValue, timestamp = null) {
const ts = timestamp || Date.now();
if (ts > this.timestamp ||
(ts === this.timestamp && this.nodeId > this.getLastWriter())) {
const oldValue = this.value;
this.value = newValue;
this.timestamp = ts;
this.vectorClock.increment();
this.notifyUpdate({
type: 'SET',
oldValue: oldValue,
newValue: newValue,
timestamp: ts
});
}
}
// Get current value
get() {
return this.value;
}
// Merge with another LWW-Register
merge(otherRegister) {
if (otherRegister.timestamp > this.timestamp ||
(otherRegister.timestamp === this.timestamp &&
otherRegister.nodeId > this.nodeId)) {
const oldValue = this.value;
this.value = otherRegister.value;
this.timestamp = otherRegister.timestamp;
this.notifyUpdate({
type: 'MERGE',
oldValue: oldValue,
newValue: this.value,
mergedFrom: otherRegister
});
}
// Merge vector clocks
this.vectorClock.merge(otherRegister.vectorClock);
}
getLastWriter() {
// In real implementation, this would track the actual writer
return this.nodeId;
}
onUpdate(callback) {
this.updateCallbacks.push(callback);
}
notifyUpdate(delta) {
this.updateCallbacks.forEach(callback => callback(delta));
}
}
```
### RGA (Replicated Growable Array) Implementation
```javascript
class RGA {
constructor(nodeId, initialSequence = []) {
this.nodeId = nodeId;
this.sequence = [];
this.tombstones = new Set();
this.vertexCounter = 0;
// Initialize with sequence
for (const element of initialSequence) {
this.insert(this.sequence.length, element);
}
this.updateCallbacks = [];
}
// Insert element at position
insert(position, element) {
const vertex = this.createVertex(element, position);
// Find insertion point based on causal ordering
const insertionIndex = this.findInsertionIndex(vertex, position);
this.sequence.splice(insertionIndex, 0, vertex);
this.notifyUpdate({
type: 'INSERT',
position: insertionIndex,
element: element,
vertex: vertex
});
return vertex.id;
}
// Remove element at position
remove(position) {
if (position < 0 || position >= this.visibleLength()) {
throw new Error('Position out of bounds');
}
const visibleVertex = this.getVisibleVertex(position);
if (visibleVertex) {
this.tombstones.add(visibleVertex.id);
this.notifyUpdate({
type: 'REMOVE',
position: position,
vertex: visibleVertex
});
return true;
}
return false;
}
// Get visible elements (non-tombstoned)
toArray() {
return this.sequence
.filter(vertex => !this.tombstones.has(vertex.id))
.map(vertex => vertex.element);
}
// Get visible length
visibleLength() {
return this.sequence.filter(vertex => !this.tombstones.has(vertex.id)).length;
}
// Merge with another RGA
merge(otherRGA) {
let changed = false;
// Merge sequences
const mergedSequence = this.mergeSequences(this.sequence, otherRGA.sequence);
if (mergedSequence.length !== this.sequence.length) {
this.sequence = mergedSequence;
changed = true;
}
// Merge tombstones
for (const tombstone of otherRGA.tombstones) {
if (!this.tombstones.has(tombstone)) {
this.tombstones.add(tombstone);
changed = true;
}
}
if (changed) {
this.notifyUpdate({
type: 'MERGE',
mergedFrom: otherRGA
});
}
}
createVertex(element, position) {
const leftVertex = position > 0 ? this.getVisibleVertex(position - 1) : null;
return {
id: `${this.nodeId}-${++this.vertexCounter}`,
element: element,
leftOrigin: leftVertex ? leftVertex.id : null,
timestamp: Date.now(),
nodeId: this.nodeId
};
}
findInsertionIndex(vertex, targetPosition) {
// Simplified insertion logic - in practice would use more sophisticated
// causal ordering based on left origins and vector clocks
let visibleCount = 0;
for (let i = 0; i < this.sequence.length; i++) {
if (!this.tombstones.has(this.sequence[i].id)) {
if (visibleCount === targetPosition) {
return i;
}
visibleCount++;
}
}
return this.sequence.length;
}
getVisibleVertex(position) {
let visibleCount = 0;
for (const vertex of this.sequence) {
if (!this.tombstones.has(vertex.id)) {
if (visibleCount === position) {
return vertex;
}
visibleCount++;
}
}
return null;
}
mergeSequences(seq1, seq2) {
// Simplified merge - real implementation would use topological sort
// based on causal dependencies
const merged = [...seq1];
for (const vertex of seq2) {
if (!merged.find(v => v.id === vertex.id)) {
merged.push(vertex);
}
}
// Sort by timestamp for basic ordering
return merged.sort((a, b) => a.timestamp - b.timestamp);
}
onUpdate(callback) {
this.updateCallbacks.push(callback);
}
notifyUpdate(delta) {
this.updateCallbacks.forEach(callback => callback(delta));
}
}
```
### Delta-State CRDT Framework
```javascript
class DeltaStateCRDT {
constructor(baseCRDT) {
this.baseCRDT = baseCRDT;
this.deltaBuffer = [];
this.lastSyncVector = new Map();
this.maxDeltaBuffer = 1000;
}
// Apply operation and track delta
applyOperation(operation) {
const oldState = this.baseCRDT.clone();
const result = this.baseCRDT.applyOperation(operation);
const newState = this.baseCRDT.clone();
// Compute delta
const delta = this.computeDelta(oldState, newState);
this.addDelta(delta);
return result;
}
// Add delta to buffer
addDelta(delta) {
this.deltaBuffer.push({
delta: delta,
timestamp: Date.now(),
vectorClock: this.baseCRDT.vectorClock.clone()
});
// Maintain buffer size
if (this.deltaBuffer.length > this.maxDeltaBuffer) {
this.deltaBuffer.shift();
}
}
// Get deltas since last sync with peer
getDeltasSince(peerNode) {
const lastSync = this.lastSyncVector.get(peerNode) || new VectorClock();
return this.deltaBuffer.filter(deltaEntry =>
deltaEntry.vectorClock.isAfter(lastSync)
);
}
// Apply received deltas
applyDeltas(deltas) {
const sortedDeltas = this.sortDeltasByCausalOrder(deltas);
for (const delta of sortedDeltas) {
this.baseCRDT.merge(delta.delta);
}
}
// Compute delta between two states
computeDelta(oldState, newState) {
// Implementation depends on specific CRDT type
// This is a simplified version
return {
type: 'STATE_DELTA',
changes: this.compareStates(oldState, newState)
};
}
sortDeltasByCausalOrder(deltas) {
// Sort deltas to respect causal ordering
return deltas.sort((a, b) => {
if (a.vectorClock.isBefore(b.vectorClock)) return -1;
if (b.vectorClock.isBefore(a.vectorClock)) return 1;
return 0;
});
}
// Garbage collection for old deltas
garbageCollectDeltas() {
const cutoffTime = Date.now() - (24 * 60 * 60 * 1000); // 24 hours
this.deltaBuffer = this.deltaBuffer.filter(
deltaEntry => deltaEntry.timestamp > cutoffTime
);
}
}
```
## MCP Integration Hooks
### Memory Coordination for CRDT State
```javascript
// Store CRDT state persistently
await this.mcpTools.memory_usage({
action: 'store',
key: `crdt_state_${this.crdtName}`,
value: JSON.stringify({
type: this.crdtType,
state: this.serializeState(),
vectorClock: Array.from(this.vectorClock.entries()),
lastSync: Array.from(this.lastSyncVector.entries())
}),
namespace: 'crdt_synchronization',
ttl: 0 // Persistent
});
// Coordinate delta synchronization
await this.mcpTools.memory_usage({
action: 'store',
key: `deltas_${this.nodeId}_${Date.now()}`,
value: JSON.stringify(this.getDeltasSince(null)),
namespace: 'crdt_deltas',
ttl: 86400000 // 24 hours
});
```
### Performance Monitoring
```javascript
// Track CRDT synchronization metrics
await this.mcpTools.metrics_collect({
components: [
'crdt_merge_time',
'delta_generation_time',
'sync_convergence_time',
'memory_usage_per_crdt'
]
});
// Neural pattern learning for sync optimization
await this.mcpTools.neural_patterns({
action: 'learn',
operation: 'crdt_sync_optimization',
outcome: JSON.stringify({
syncPattern: this.lastSyncPattern,
convergenceTime: this.lastConvergenceTime,
networkTopology: this.networkState
})
});
```
## Advanced CRDT Features
### Causal Consistency Tracker
```javascript
class CausalTracker {
constructor(nodeId) {
this.nodeId = nodeId;
this.vectorClock = new VectorClock(nodeId);
this.causalBuffer = new Map();
this.deliveredEvents = new Set();
}
// Track causal dependencies
trackEvent(event) {
event.vectorClock = this.vectorClock.clone();
this.vectorClock.increment();
// Check if event can be delivered
if (this.canDeliver(event)) {
this.deliverEvent(event);
this.checkBufferedEvents();
} else {
this.bufferEvent(event);
}
}
canDeliver(event) {
// Event can be delivered if all its causal dependencies are satisfied
for (const [nodeId, clock] of event.vectorClock.entries()) {
if (nodeId === event.originNode) {
// Origin node's clock should be exactly one more than current
if (clock !== this.vectorClock.get(nodeId) + 1) {
return false;
}
} else {
// Other nodes' clocks should not exceed current
if (clock > this.vectorClock.get(nodeId)) {
return false;
}
}
}
return true;
}
deliverEvent(event) {
if (!this.deliveredEvents.has(event.id)) {
// Update vector clock
this.vectorClock.merge(event.vectorClock);
// Mark as delivered
this.deliveredEvents.add(event.id);
// Apply event to CRDT
this.applyCRDTOperation(event);
}
}
bufferEvent(event) {
if (!this.causalBuffer.has(event.id)) {
this.causalBuffer.set(event.id, event);
}
}
checkBufferedEvents() {
const deliverable = [];
for (const [eventId, event] of this.causalBuffer) {
if (this.canDeliver(event)) {
deliverable.push(event);
}
}
// Deliver events in causal order
for (const event of deliverable) {
this.causalBuffer.delete(event.id);
this.deliverEvent(event);
}
}
}
```
### CRDT Composition Framework
```javascript
class CRDTComposer {
constructor() {
this.compositeTypes = new Map();
this.transformations = new Map();
}
// Define composite CRDT structure
defineComposite(name, schema) {
this.compositeTypes.set(name, {
schema: schema,
factory: (nodeId, replicationGroup) =>
this.createComposite(schema, nodeId, replicationGroup)
});
}
createComposite(schema, nodeId, replicationGroup) {
const composite = new CompositeCRDT(nodeId, replicationGroup);
for (const [fieldName, fieldSpec] of Object.entries(schema)) {
const fieldCRDT = this.createFieldCRDT(fieldSpec, nodeId, replicationGroup);
composite.addField(fieldName, fieldCRDT);
}
return composite;
}
createFieldCRDT(fieldSpec, nodeId, replicationGroup) {
switch (fieldSpec.type) {
case 'counter':
return fieldSpec.decrements ?
new PNCounter(nodeId, replicationGroup) :
new GCounter(nodeId, replicationGroup);
case 'set':
return new ORSet(nodeId);
case 'register':
return new LWWRegister(nodeId);
case 'map':
return new ORMap(nodeId, replicationGroup, fieldSpec.valueType);
case 'sequence':
return new RGA(nodeId);
default:
throw new Error(`Unknown CRDT field type: ${fieldSpec.type}`);
}
}
}
class CompositeCRDT {
constructor(nodeId, replicationGroup) {
this.nodeId = nodeId;
this.replicationGroup = replicationGroup;
this.fields = new Map();
this.updateCallbacks = [];
}
addField(name, crdt) {
this.fields.set(name, crdt);
// Subscribe to field updates
crdt.onUpdate((delta) => {
this.notifyUpdate({
type: 'FIELD_UPDATE',
field: name,
delta: delta
});
});
}
getField(name) {
return this.fields.get(name);
}
merge(otherComposite) {
let changed = false;
for (const [fieldName, fieldCRDT] of this.fields) {
const otherField = otherComposite.fields.get(fieldName);
if (otherField) {
const oldState = fieldCRDT.clone();
fieldCRDT.merge(otherField);
if (!this.statesEqual(oldState, fieldCRDT)) {
changed = true;
}
}
}
if (changed) {
this.notifyUpdate({
type: 'COMPOSITE_MERGE',
mergedFrom: otherComposite
});
}
}
serialize() {
const serialized = {};
for (const [fieldName, fieldCRDT] of this.fields) {
serialized[fieldName] = fieldCRDT.serialize();
}
return serialized;
}
onUpdate(callback) {
this.updateCallbacks.push(callback);
}
notifyUpdate(delta) {
this.updateCallbacks.forEach(callback => callback(delta));
}
}
```
## Integration with Consensus Protocols
### CRDT-Enhanced Consensus
```javascript
class CRDTConsensusIntegrator {
constructor(consensusProtocol, crdtSynchronizer) {
this.consensus = consensusProtocol;
this.crdt = crdtSynchronizer;
this.hybridOperations = new Map();
}
// Hybrid operation: consensus for ordering, CRDT for state
async hybridUpdate(operation) {
// Step 1: Achieve consensus on operation ordering
const consensusResult = await this.consensus.propose({
type: 'CRDT_OPERATION',
operation: operation,
timestamp: Date.now()
});
if (consensusResult.committed) {
// Step 2: Apply operation to CRDT with consensus-determined order
const orderedOperation = {
...operation,
consensusIndex: consensusResult.index,
globalTimestamp: consensusResult.timestamp
};
await this.crdt.applyOrderedOperation(orderedOperation);
return {
success: true,
consensusIndex: consensusResult.index,
crdtState: this.crdt.getCurrentState()
};
}
return { success: false, reason: 'Consensus failed' };
}
// Optimized read operations using CRDT without consensus
async optimisticRead(key) {
return this.crdt.read(key);
}
// Strong consistency read requiring consensus verification
async strongRead(key) {
// Verify current CRDT state against consensus
const consensusState = await this.consensus.getCommittedState();
const crdtState = this.crdt.getCurrentState();
if (this.statesConsistent(consensusState, crdtState)) {
return this.crdt.read(key);
} else {
// Reconcile states before read
await this.reconcileStates(consensusState, crdtState);
return this.crdt.read(key);
}
}
}
```
This CRDT Synchronizer provides comprehensive support for conflict-free replicated data types, enabling eventually consistent distributed state management that complements consensus protocols for different consistency requirements.Проверка и особенности
Автоматическая проверка источника и файлов
Автоматически сверены публичный репозиторий, лицензия, YAML-заголовок SKILL.md, состав папки и Git-хеш каждого файла. Исполняемые файлы и срабатывания статических правил исключены из этой партии. Ручной аудит всех инструкций и проверка работы не проводились; эта проверка не гарантирует безопасность. Команда закреплена на проверенном коммите.
Интерес к репозиторию оцениваем по звёздам и форкам GitHub. Это показатели всего репозитория, а не числа установок отдельного скилла.
- Проверка состава
- 16.09.2026
- Проверенная версия
6f0ed7112873- Файлы с кодом
- Не указаны в данных проверки
- Изменения на GitHub
- Нужна повторная проверкаПоследняя проверка: 11.10.2026
Репозиторий и лицензия
Это показатели всего репозитория, а не установки скилла. Снимок требует обновления.
- Последняя активность GitHub
- 11.10.2026
- Лицензия
- MIT
- Добавлен в каталог
- 16.09.2026
Полный текст лицензии
MIT License Copyright (c) 2024-2026 ruvnet Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
Ещё в этой категории
Все в категорииОбсуждение
ПравилаПользовались этим скиллом? Делитесь опытом без регистрации. Все комментарии проверяет модератор.
