Data Sync

Data Sync

Run or inspect data synchronization workflows for: $ARGUMENTS

Overview

• Implementation status: workflow-only
• Current backing path: use runtime artifacts, manifests, bundle diagnostics, and host-provided sync steps
• Primary purpose: define the data-refresh, source-health, and freshness-audit workflow without overstating local orchestration support
• Research layer: data infrastructure (Stage 2: Data Collection & Quality Assurance - Data refresh and validation)
• Workflow stage: stage 2 Data Collection & Quality Assurance

Use When

• The user wants to refresh source data.
• The user wants to compare provider behavior or diagnose stale records.
• The caller needs an explicit audit of whether current artifacts are fresh enough for research, screening, or backtesting.
• The user wants to know what must happen before a clean rerun.
• The user wants to verify data freshness before critical decisions.
• The user wants to diagnose data quality issues or provider failures.

Do Not Use When

• The user wants immediate one-symbol analysis and the existing data is already sufficient.
• The user only wants runtime context locations. Use session-status.
• The user wants normalization or feature enrichment of an already available dataset. Use market-data.
• The user wants to analyze data quality after collection. Use market-data for quality assessment.

Inputs

• A symbol, a provider name, a date window, or a sync scope.
• Optional host-side parameters such as freshness threshold, source priority, or required artifact types.
• Optional user requirement such as:
• current-day screening freshness
• backtest-grade historical coverage
• point-in-time fundamental correctness
• explicit provider comparison

Execution

Step 1: Classify the sync request

Determine which type of sync operation is needed:

Sync request types:

• Freshness audit: Check if existing data is recent enough for intended use
• Provider comparison: Compare data from multiple sources for consistency
• Missing history diagnosis: Identify gaps in historical data coverage
• Symbol-level refresh: Update data for specific symbols
• Artifact lineage check: Verify data provenance and transformation chain
• Pre-backtest readiness check: Validate data quality for backtesting
• Bulk refresh: Update entire universe or watchlist
• Incremental update: Add only new data since last sync

Use case classification:

• Descriptive analysis: Lower freshness requirement (1-2 days stale acceptable)
• Screening: Medium freshness requirement (same-day data preferred)
• Decision support: High freshness requirement (intraday data needed)
• Backtesting: Historical completeness required (no gaps, point-in-time correct)
• Live trading: Real-time or near-real-time data required

Step 2: Inspect existing data state

Audit current data availability and quality:

Data inventory:

• [ ] Which symbols have data available
• [ ] Date range for each symbol (start date, end date)
• [ ] Data freshness (latest date, days since last update)
• [ ] Data source (provider, file, cache)
• [ ] Data completeness (% of expected dates present)
• [ ] Data quality score (from market-data if available)

Artifact inspection:

• [ ] Runtime artifacts (state.json, report.md, metadata.json)
• [ ] Manifests (data source declarations, version info)
• [ ] Bundle diagnostics (error logs, sync logs)
• [ ] File timestamps (when files were last modified)
• [ ] Cache status (what is cached, cache age)

Provider status:

• [ ] Which providers are configured (akshare, tushare, etc.)
• [ ] Provider health (last successful fetch, error rate)
• [ ] Provider rate limits (requests remaining, reset time)
• [ ] Provider coverage (which symbols, which fields)
• [ ] Provider latency (typical fetch time)

Step 3: Evaluate data readiness

Assess whether existing data meets requirements:

Freshness evaluation:

• Fresh ( 7 days old): Requires refresh before any use
• Missing: No data available, must fetch

Completeness evaluation:

• Complete (100%): All expected dates present
• Mostly complete (90-99%): Minor gaps, acceptable for most uses
• Incomplete (70-89%): Significant gaps, use with caution
• Sparse (< 70%): Major gaps, unsuitable for analysis

Quality evaluation:

• High quality (score 90-100): Suitable for all uses including backtesting
• Good quality (score 70-89): Suitable for descriptive analysis, screening
• Fair quality (score 50-69): Use with caution, validate results
• Poor quality (score < 50): Requires refresh or alternative source

Point-in-time correctness (for backtesting):

• [ ] Corporate actions properly adjusted (splits, dividends)
• [ ] No look-ahead bias (data as-of historical dates)
• [ ] Survivorship bias addressed (delisted symbols included)
• [ ] Restatements handled (as-reported vs. as-restated)
• [ ] Suspension periods identified

Step 4: Diagnose data issues

Identify specific problems requiring attention:

Common data issues:

• Stale data: Latest date is too old for intended use
• Missing symbols: Symbols in watchlist have no data
• Incomplete history: Gaps in date coverage
• Provider failures: Recent fetch attempts failed
• Quality degradation: Data quality score declining over time
• Inconsistent sources: Different providers show different values
• Corporate action errors: Splits/dividends not properly adjusted
• Suspension gaps: Suspension periods not identified
• Delisting issues: Delisted symbols missing from historical data

For each issue:

• Issue type and severity (critical, high, medium, low)
• Affected symbols (count and list)
• Impact on downstream workflows (which skills affected)
• Recommended action (refresh, validate, alternative source)
• Automation status (can be fixed automatically or requires manual intervention)

Step 5: Generate sync plan

Create actionable plan to address issues:

Sync plan components:

Part 1: Immediate actions (critical issues):

• Symbols requiring immediate refresh (stale data blocking decisions)
• Provider health checks (if recent failures detected)
• Cache invalidation (if corrupted data suspected)
• Manual interventions (if automation unavailable)

Part 2: Scheduled actions (high priority):

• Bulk refresh for watchlist (if many symbols stale)
• Historical backfill (if gaps detected)
• Provider comparison (if inconsistencies suspected)
• Quality validation (if quality scores declining)

Part 3: Maintenance actions (medium priority):

• Routine refresh schedule (daily, weekly)
• Cache cleanup (remove old artifacts)
• Provider rotation (switch to backup if primary failing)
• Monitoring setup (alerts for future staleness)

Part 4: Validation actions (before critical use):

• Pre-backtest validation (point-in-time correctness)
• Pre-decision validation (freshness and quality)
• Cross-provider validation (consistency check)
• Manual spot checks (sample verification)

Step 6: Execute or delegate sync operations

Perform sync based on automation availability:

If local automation available:

• Execute refresh commands directly
• Monitor progress and handle errors
• Validate results after sync
• Update artifacts and manifests

If host-framework automation available:

• Delegate to host-framework sync API
• Provide sync parameters (symbols, date range, providers)
• Monitor sync status
• Validate results after sync

If manual intervention required:

• Provide explicit instructions for manual refresh
• List specific commands to run
• Specify validation steps after manual sync
• Document what was done for audit trail

Sync execution priorities:

1. Critical symbols (blocking immediate decisions)
2. High-priority watchlist (screening, monitoring)
3. Historical backfill (backtesting preparation)
4. Routine maintenance (scheduled updates)

Step 7: Validate sync results

After sync, verify data quality:

Post-sync validation:

• [ ] Freshness improved (latest date is now recent)
• [ ] Completeness improved (gaps filled)
• [ ] Quality maintained or improved (quality score)
• [ ] No new issues introduced (consistency checks)
• [ ] Artifacts updated (manifests, metadata)

Validation checks:

• Compare before/after freshness
• Compare before/after completeness
• Compare before/after quality scores
• Check for new gaps or inconsistencies
• Verify corporate action adjustments

If validation fails:

• Identify specific failures
• Recommend retry with different provider
• Recommend manual intervention
• Document issue for future investigation

Step 8: Generate sync report

Organize findings into structured report:

Part 1: Sync Summary

• Sync scope (symbols, date range, providers)
• Sync timestamp
• Sync status (success, partial, failed)
• Symbols refreshed (count and list)
• Issues resolved (count and list)

Part 2: Data Inventory

• Total symbols with data
• Freshness distribution (fresh, recent, stale, very stale, missing)
• Completeness distribution (complete, mostly complete, incomplete, sparse)
• Quality distribution (high, good, fair, poor)
• Provider distribution (which providers used)

Part 3: Issues Diagnosed For each issue:

• Issue type and severity
• Affected symbols
• Impact on workflows
• Recommended action
• Automation status

Part 4: Sync Plan

• Immediate actions (critical)
• Scheduled actions (high priority)
• Maintenance actions (medium priority)
• Validation actions (before critical use)

Part 5: Sync Execution Results

• Actions taken (refresh, backfill, validation)
• Success rate (% of actions successful)
• Failures (count and reasons)
• Manual interventions required

Part 6: Post-Sync Validation

• Freshness improvement (before/after)
• Completeness improvement (before/after)
• Quality improvement (before/after)
• Remaining issues (if any)

Part 7: Readiness Assessment

• Ready for descriptive analysis? (Yes/No/Conditional)
• Ready for screening? (Yes/No/Conditional)
• Ready for decision support? (Yes/No/Conditional)
• Ready for backtesting? (Yes/No/Conditional)
• Conditions or caveats (if conditional)

Part 8: Next Steps

• Downstream workflows ready to proceed
• Workflows blocked pending further sync
• Monitoring and alerting recommendations
• Scheduled maintenance recommendations

Step 9: Return explicit sync diagnosis

When delivering results, maintain proper framing:

Sync interpretation:

• This is data infrastructure status, not investment recommendation
• Freshness and quality assessments are technical, not fundamental
• Readiness evaluation is for workflow purposes, not market timing
• Sync recommendations are operational, not strategic

Automation transparency:

• State which sync operations are locally automated
• State which operations require host-framework support
• State which operations require manual intervention
• State limitations of current automation

Readiness framing:

• Separate readiness for different use cases (analysis, screening, backtesting)
• State specific conditions or caveats for conditional readiness
• Recommend validation steps before critical use
• Tie readiness to next workflow stage

Limitation disclosure:

• State what sync can and cannot guarantee
• State point-in-time correctness limitations
• State survivorship bias limitations
• State provider coverage limitations

Output Contract

• Expected result: sync status, stale-data diagnosis, provider comparison, or an explicit next-action plan.
• Caller-facing delivery standard:
• Eight-part structure: Sync summary, data inventory, issues diagnosed, sync plan, execution results, post-sync validation, readiness assessment, next steps
• Scope identification: State what is being audited (symbols, date range, providers)
• Evidence basis: Identify evidence used (manifests, timestamps, runtime status, file inspection)
• Automation transparency: State whether locally automated, partially automated, or host-dependent
• Readiness assessment: Tie sync recommendation to next workflow stage (screening, analysis, backtesting, trading)
• Issue diagnosis: Specific issues with severity, affected symbols, impact, recommended actions
• Sync plan: Immediate, scheduled, maintenance, and validation actions
• Validation results: Before/after comparison of freshness, completeness, quality
• Limitation disclosure: State what sync can and cannot guarantee (point-in-time, survivorship, etc.)
• Local limitation:
• The skill does not currently guarantee a full local provider-orchestration layer
• The workflow may depend on external scheduling or host-side refresh behavior
• Some sync operations may require manual intervention

Failure Handling

• If the requested sync scope is not supported locally, state that explicitly.
• If available evidence is insufficient to diagnose freshness, say what is missing.
• If the user asks for backtest-grade readiness and the skill cannot verify survivorship, corporate-action, or point-in-time integrity, say so directly.
• If provider is unavailable or failing, recommend alternative sources or manual intervention.
• If sync execution fails, provide specific error details and retry recommendations.

Key Rules

• Do not imply a full provider-orchestration layer where none exists locally.
• Separate source-health diagnostics from market-analysis claims.
• Prefer explicit freshness and readiness language over vague advice to “rerun.”
• Treat descriptive-analysis readiness as a lower bar than backtest or live-verification readiness.
• Automation transparency is mandatory. State what is automated vs. manual.
• Readiness must be use-case specific. Different standards for analysis, screening, backtesting, trading.
• Issue diagnosis must be specific. Type, severity, affected symbols, impact, actions.
• Sync plan must be actionable. Specific commands, parameters, validation steps.
• Validation must be quantitative. Before/after metrics, not just “looks better.”
• Limitation disclosure is mandatory. State what sync cannot guarantee.

Composition

• Often paired with session-status, market-data, stock-data, and backtest-evaluator.
• Serves as the stage-2 control surface before broader research or validation workflows.
• Should be run before market-screen or decision-dashboard if data freshness is uncertain.
• Should be run before backtest-evaluator to ensure historical data quality.
• Can be triggered by market-data if quality issues detected.
• Results feed into readiness decisions for all downstream skills.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: NatsuFox
• Source: NatsuFox/A-Stockit
• License: MIT

Install and usage instructions live in the source repository linked above.