synchronizeSafety 85Repository ShareFavorite skill

Intent

Execute synchronize to merge data from multiple sources into a coherent, unified result with explicit conflict resolution.

Success criteria:

• All source data is accounted for (merged, replaced, or explicitly dropped)
• Conflicts are detected and resolved according to specified strategy
• Provenance is preserved for each merged element
• Output includes lineage trace for every change

Compatible schemas:

• schemas/output_schema.yaml

Inputs

Procedure

1. Inventory sources: List all sources to synchronize

   • Record source type (agent output, file, API response)
   • Extract timestamp or version if available
   • Note source authority level (primary, secondary, derived)

2. Normalize structures: Transform all sources to comparable format

   • Identify common keys/fields across sources
   • Map equivalent fields with different names
   • Flag structural incompatibilities

3. Detect conflicts: Compare values for overlapping entities

   • Exact match: no conflict
   • Value difference: record as conflict with both values
   • Missing in one source: record as addition/deletion

4. Apply resolution strategy:

   • merge: Combine non-conflicting data; use conflict_rules for overlaps
   • replace: Use most recent or highest-priority source
   • conflict_resolve: Apply explicit rules; flag unresolvable for human review

5. Ground claims: Attach provenance to each merged element

   • Format: source:<source_id>:<path> or conflict:<resolution_method>
   • Track which source contributed each field

6. Validate output: Check merged result against target schema if provided

7. Format output: Structure results according to the output contract below

Output Contract

Return a structured object:

synchronization:
  sources: array[string]  # IDs of sources synchronized
  strategy: merge | replace | conflict_resolve
  timestamp: string  # ISO timestamp of sync operation
changes:
  - source: string  # Which source provided this change
    type: add | modify | delete
    entity: string  # Entity ID or path affected
    field: string | null  # Specific field if applicable
    value: object  # New/changed value
conflicts:
  - sources: array[string]  # Sources with conflicting values
    attribute: string  # Field/path in conflict
    values: array[object]  # Conflicting values from each source
    resolution: string | null  # How resolved, or null if unresolved
    resolution_rationale: string | null
result:
  synchronized: boolean  # True if all conflicts resolved
  entities_affected: integer  # Count of entities changed
  unresolved_conflicts: integer  # Count of conflicts needing human review
merged_data: object  # The synchronized result
confidence: number  # 0.0-1.0 based on conflict count and resolution quality
evidence_anchors: array[string]  # Source references for each claim
assumptions: array[string]  # Explicit assumptions made during sync

Field Definitions

Examples

Example 1: Merging Two Agent Outputs

Input:

sources:
  - agent_output_1:
      entity: "user_123"
      name: "Alice"
      role: "admin"
      last_seen: "2024-01-15"
  - agent_output_2:
      entity: "user_123"
      name: "Alice Smith"
      department: "Engineering"
      last_seen: "2024-01-16"
strategy: merge
conflict_rules:
  name: prefer_longer
  last_seen: prefer_newer

Output:

synchronization:
  sources: ["agent_output_1", "agent_output_2"]
  strategy: merge
  timestamp: "2024-01-16T14:30:00Z"
changes:
  - source: "agent_output_1"
    type: add
    entity: "user_123"
    field: "role"
    value: "admin"
  - source: "agent_output_2"
    type: add
    entity: "user_123"
    field: "department"
    value: "Engineering"
  - source: "agent_output_2"
    type: modify
    entity: "user_123"
    field: "name"
    value: "Alice Smith"
  - source: "agent_output_2"
    type: modify
    entity: "user_123"
    field: "last_seen"
    value: "2024-01-16"
conflicts:
  - sources: ["agent_output_1", "agent_output_2"]
    attribute: "name"
    values: ["Alice", "Alice Smith"]
    resolution: "prefer_longer"
    resolution_rationale: "Longer name likely more complete"
  - sources: ["agent_output_1", "agent_output_2"]
    attribute: "last_seen"
    values: ["2024-01-15", "2024-01-16"]
    resolution: "prefer_newer"
    resolution_rationale: "More recent timestamp preferred"
result:
  synchronized: true
  entities_affected: 1
  unresolved_conflicts: 0
merged_data:
  entity: "user_123"
  name: "Alice Smith"
  role: "admin"
  department: "Engineering"
  last_seen: "2024-01-16"
confidence: 0.95
evidence_anchors:
  - "source:agent_output_1:role"
  - "source:agent_output_2:department"
  - "conflict:name:prefer_longer"
  - "conflict:last_seen:prefer_newer"
assumptions:
  - "Both sources refer to same user_123 entity"
  - "Longer name is more complete (not a different person)"

Evidence pattern: Each field in merged_data traced to originating source; conflicts logged with resolution method.

Example 2: Unresolvable Conflict

Input:

sources:
  - system_a: { price: 100, currency: "USD" }
  - system_b: { price: 150, currency: "USD" }
strategy: conflict_resolve
conflict_rules:
  price: require_human_review

Output:

synchronization:
  sources: ["system_a", "system_b"]
  strategy: conflict_resolve
  timestamp: "2024-01-16T15:00:00Z"
changes: []
conflicts:
  - sources: ["system_a", "system_b"]
    attribute: "price"
    values: [100, 150]
    resolution: null
    resolution_rationale: "Price conflict requires human review per policy"
result:
  synchronized: false
  entities_affected: 0
  unresolved_conflicts: 1
merged_data: null
confidence: 0.3
evidence_anchors:
  - "source:system_a:price"
  - "source:system_b:price"
assumptions:
  - "Both prices refer to same product"
  - "50% price discrepancy exceeds auto-resolve threshold"

Verification

• All source data accounted for in changes or conflicts
• Every conflict has resolution or is flagged unresolved
• merged_data passes target_schema validation (if provided)
• confidence reflects conflict severity and resolution quality
• evidence_anchors trace every merged field to source

Verification tools: Read (to validate source accessibility), Grep (to search for schema definitions)

Safety Constraints

• mutation: false
• requires_checkpoint: false
• requires_approval: false
• risk: low

Capability-specific rules:

• Do not auto-resolve conflicts involving security-sensitive fields (credentials, permissions)
• Preserve original source data in evidence_anchors; never discard
• Flag confidence < 0.5 if more than 30% of fields have conflicts
• Stop and request clarification if sources have incompatible schemas

Composition Patterns

Commonly follows:

• retrieve - Gathers raw data from multiple sources for synchronization
• transform - Normalizes data formats before merge
• receive - Ingests incoming events to synchronize with existing state
• delegate - Returns outputs from parallel agents to be merged

Commonly precedes:

• world-state - Synchronized data becomes canonical state snapshot
• identity-resolution - Merged entities may need identity reconciliation
• persist - Save synchronized result to durable storage
• verify - Validate merged output meets requirements

Anti-patterns:

• Never synchronize without grounding - merge requires evidence anchors (see ontology edge)
• Never synchronize without provenance - merge requires lineage trace (see ontology edge)
• Avoid synchronizing security-sensitive data without human review

Workflow references:

• See reference/composition_patterns.md#digital-twin-sync-loop for twin state synchronization
• See reference/composition_patterns.md#enrichment-pipeline for parallel data gathering