database-manager（数据库管理器）

Setup

On first use, read setup.md for local initialization, activation preferences, and operating defaults.

When to Use

User needs a reliable database operating system for schema design, query hygiene, migration rollout, and recovery readiness. Agent keeps data work safe by enforcing preflight checks, explicit rollback plans, and incident-ready runbooks.

Use this skill when database changes can affect production reliability, latency, or data integrity.

Architecture

Memory lives in ~/database-manager/. See memory-template.md for the base structure.

~/database-manager/
├── memory.md                  # Durable context and operating preferences
├── inventory.md               # Systems, engines, owners, and critical datasets
├── standards.md               # Naming, indexing, and schema conventions
├── migrations.md              # Planned and executed migration records
├── backups.md                 # Backup schedule, retention, and restore drills
├── incidents.md               # Incident timeline, mitigations, and follow-up
└── archive/
    ├── migrations-YYYY-MM.md  # Closed migrations by month
    └── incidents-YYYY-MM.md   # Closed incidents by month

Quick Reference

Core Rules

1. Never Change Production Without a Preflight Packet

Every schema or data change must include:

• intent
• blast radius
• rollback path
• verification query set

Skipping preflight creates unbounded operational risk.

2. Separate Read Validation from Write Execution

Validate assumptions with read-only checks first, then run writes in explicit, audited steps.

This prevents accidental broad updates caused by stale predicates or wrong join keys.

3. Treat Migrations as Product Releases

Each migration requires:

• owner
• deployment window
• rollback deadline
• post-deploy verification criteria

Schema changes without release discipline are a primary source of prolonged incidents.

4. Make Index and Query Trade-offs Explicit

When changing indexes or query plans, state expected impact on:

• read latency
• write throughput
• storage growth

Invisible trade-offs create hidden cost and unpredictable performance regressions.

5. Backup Is Not Real Until Restore Is Proven

Do not trust backup status alone. Run restore drills, validate recovered row counts, and document measured recovery time.

Unverified backups are operationally equivalent to no backups.

6. Encode Safety Gates for Destructive Operations

Before DROP, TRUNCATE, broad DELETE, or bulk UPDATE:

• confirm table and environment
• capture row count baseline
• require explicit user confirmation
• log exact rollback route

Destructive steps without safety gates can permanently corrupt business data.

7. Close Every Incident with Durable Learning

After mitigation:

• capture root cause
• record missing guardrail
• add one concrete prevention change

Without closure rules, the same incident class repeats.

Common Traps

• Running updates without a restrictive predicate -> unintended mass writes and long rollback windows.
• Shipping migrations without timing and lock analysis -> production latency spikes and blocked transactions.
• Assuming replica lag is harmless -> stale reads and misleading verification outcomes.
• Declaring backup success without restore drills -> false confidence during outages.
• Fixing incidents without permanent guardrails -> repeated operational failures.

Security & Privacy

Data that leaves your machine:

• None by default.

Data that stays local:

• Database operating context and records under ~/database-manager/.

This skill does NOT:

• Execute destructive commands without explicit user confirmation.
• Access unrelated credentials or services by default.
• Store secrets in memory files.

Related Skills

Install with clawhub install <slug> if user confirms:

• sql - SQL query authoring and analysis
• mysql - MySQL-specific workflows and troubleshooting
• prisma - Prisma schema and migration tooling
• sqlite - local database workflows and prototyping
• backend - backend architecture and service delivery

Feedback

• If useful: clawhub star database-manager
• Stay updated: clawhub sync