Raw data storage layer for 1ai-skills. Provides structured data persistence, query interface, and data pipeline support for skill operations. history. Use when working with data.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
npx skills add https://github.com/oyi77/1ai-skills --skill data
The command stays on one line. Scroll horizontally to inspect it before copying.
Prefer a local copy? Download the files currently available to SkillsMP.
Raw data storage layer for 1ai-skills. Provides structured data persistence, query interface, and data pipeline support for skill operations. history. Use when working with data.
defavg_latency_by_skill(skill_name: str, days: int = 7) -> float | None:
"""Return average execution latency in ms for a skill over N days."""
conn = sqlite3.connect(_expand_path(DB_PATH))
try:
row = conn.execute("""
SELECT AVG(latency_ms) FROM skill_executions
WHERE skill_name = ?
AND timestamp >= datetime('now', ?)
""", (skill_name, f"-{days} days")).fetchone()
returnround(row[0], 2) if row and row[0] elseNonefinally:
conn.close()
# p50 = avg_latency_by_skill("seo-optimizer")# print(f"7-day avg latency: {p50}ms")
Retrieving improvement candidates with high impact
defget_candidates(min_impact: float = 0.7, status: str = "proposed"):
"""Fetch improvement candidates above a minimum impact score."""
conn = sqlite3.connect(_expand_path(DB_PATH))
conn.row_factory = sqlite3.Row
try:
rows = conn.execute("""
SELECT skill_name, title, impact_score, effort_score,
status, created_at
FROM improvement_candidates
WHERE impact_score >= ? AND status = ?
ORDER BY impact_score DESC
""", (min_impact, status)).fetchall()
return [dict(r) for r in rows]
finally:
conn.close()
# improvements = get_candidates(0.7, "proposed")# for c in improvements:# print(f"{c['skill_name']}: {c['title']} (impact={c['impact_score']})")
Integration
Connects to:
performance-monitor (writes metrics)
feedback-collector (stores feedback)
pattern-recognition (queries patterns)
skill-evolution (tracks versions)
When NOT to Use
When the skill is stable and not changing
For skills with fewer than 10 invocations (not enough data)
When manual curation produces better results
Overview
Data is a foundational meta-skills skill that provides skill management capabilities for the agent ecosystem.
It serves as the persistence backbone for the entire 1ai-skills ecosystem: execution metrics, feedback records, improvement proposals, and skill version histories all flow through the data skill's SQLite-backed store. Without it, the self-improvement loop that drives skill evolution would have no memory.
Storage Model
The data layer uses a hybrid architecture. A primary SQLite database holds structured records — skill execution entries, latency metrics, success/failure counts, feedback items, and improvement candidates. Each record is tagged with skill name, timestamp, and execution context for precise querying. Read-heavy dashboards and trend queries are served by materialized aggregation tables refreshed on write.
Lifecycle
Data enters through instrumentation hooks embedded in every skill's execution path. The payload is validated against a schema, enriched with session metadata, and committed transactionally. A background maintenance process periodically compacts old records, prunes data beyond the retention window (90 days for raw execution logs, 18 months for aggregated metrics), and updates summary tables. This keeps the store lean while preserving historical trends.
Workflow
Define Schema — Establish the record format for each data type (skill execution, feedback, improvement candidate). Declare columns, types, indexes, and constraints in a schema registry before any data flows.
Collect Instrumentation — Hooks at skill entry/exit points emit structured payloads: skill name, duration, success/failure, error class, input hash, and output summary. Payloads are batched and sent to the data store asynchronously.
Validate and Normalize — Incoming records are checked against schema rules. Malformed entries are rejected with a detailed error logged. Valid records are normalized — timestamps converted to UTC, enums standardized to lowercase, null fields filled with sensible defaults.
Store Transactionally — Records are inserted in a single SQLite transaction per batch. WAL journal mode allows concurrent reads during writes. Duplicate detection uses a composite key of (skill_name, timestamp, execution_id).
Index and Aggregate — After each write batch, summary tables are updated: rolling 7-day averages, p50/p95/p99 latency percentiles, and hourly/daily success rates. Indexes on (skill_name, timestamp) enable fast filtered queries.
Query and Analyze — The query layer exposes parameterized SQL and convenience methods: average latency by skill, success rate over time, top-N slowest executions, improvement impact scores, and trend direction indicators.
Archive and Prune — A scheduled maintenance job rotates raw data beyond the retention window to a compressed cold-storage archive, then deletes the source rows. Aggregated summaries are compacted into weekly/monthly rollups before the raw rows are dropped.
Architecture
Input layer — Receives and validates incoming requests
Processing layer — Core logic for skill management
Output layer — Formats and delivers results
State management — Maintains context across invocations
Database Schema
The core schema includes these tables:
skill_executions — One row per skill invocation. Columns: skill_name, timestamp, latency_ms, success (boolean), error_class, input_hash, output_summary, session_id, execution_id. Indexed on (skill_name, timestamp) for time-series range queries.
Foreign keys are enforced when the source system guarantees referential integrity. All timestamp columns use ISO-8601 text format for portability. A checksum column on each raw execution row detects corruption during archival and restore. Schema migrations use a versioned migration table with forward-only numeric IDs and a rollback script per migration.
Configuration
Set up required environment variables and paths
Configure logging level and output format
Define resource limits (memory, time, API calls)
Enable/disable features via configuration flags
Data-Specific Configuration
DATA_DB_PATH — Path to the primary SQLite database file. Default: ~/.1ai/data/store.sqlite.
DATA_RETENTION_DAYS — Days to retain raw execution logs before archival. Default: 90.
DATA_AGGREGATION_RETENTION_DAYS — Days to retain aggregated summaries. Default: 547 (18 months).
Concurrent process instances applying migrations at different versions.
Use a migration lock table with advisory lock. Make migrations idempotent. Avoid concurrent migration runs.
Corrupted database file
Unexpected power loss or filesystem error during write.
Enable PRAGMA integrity_check on startup. Maintain hourly WAL checkpoints. Keep the most recent automated backup.
UNIQUE constraint failed on insert
Duplicate execution_id or composite key collision.
Use INSERT OR IGNORE for idempotent inserts. Verify the caller generates unique execution IDs per batch.
Aggregation tables stale after write
Maintenance job runs on a fixed schedule, not after every batch.
Trigger aggregation refresh on write via a callback. Reduce the aggregation interval to 5 minutes.
Monetization
Approach
Timeframe
Description
Data Analytics Service
2-4 weeks
Offer skill execution analytics as a paid service for teams running custom skill sets. Provide dashboards for latency trends, failure rates, and improvement velocity.
Benchmark Reports
1-2 weeks
Generate comparative performance benchmarks across skill categories. Sell as one-off reports to enterprise users evaluating the ecosystem.
Managed Data Pipeline
4-8 weeks
Deploy the data storage layer as a managed cloud service with replication, automated backups, and SLA-backed availability. Charge per-skill per-month.
Anomaly Detection Add-on
3-6 weeks
Build anomaly detection on top of historical execution data — flag skills whose latency or failure rate deviates significantly from their baseline. License as an add-on module.
Migration Consulting
Per engagement
Help teams migrate from ad-hoc logging to the structured data layer. Includes schema design, data migration scripts, and integration with existing instrumentation.
Process
Preparation
Identify the data types that need persistence: execution metrics, feedback, versions, improvement candidates.
Define schema for each type: columns, types, constraints, indexes, default values.
Set up the SQLite database file and run initial migrations to create all tables.
Configure retention policies and archival paths before the first write.
Instrument the calling code with entry/exit hooks that emit structured payloads.
Execution
Route incoming data payloads through the validation layer before any write.
Use batched transactions with WAL mode for write performance.
Update aggregation tables after each batch to keep summary queries fast.
Run the query interface for analysis: latency percentiles, success rates, trend detection.
Schedule maintenance tasks (pruning, compaction, backup) via cron or the built-in scheduler.
