<< All versions

Skill v1.0.0

currentAutomated scan100/100
joao-aires/agent-skills-public/postgresql-operations
──Details
PublishedOctober 23, 2026 at 01:32 PM
Content Hashsha256:ab01de900ad8d901...
Git SHAed9171c19ba3
──Files
Files (1 file, 3.9 KB)
SKILL.md3.9 KBactive
SKILL.md · 60 lines · 3.9 KB

version: "1.0.0" name: postgresql-operations description: Plan or perform scoped PostgreSQL backup/recovery, maintenance, replication and upgrades. Use for operating procedures and readiness with the actual hosting model, recovery objectives and authorized target. license: MIT metadata: scope: stack-specific invocation: both role: support stack: PostgreSQL references: ../../references/sources.md user-invocable: true


Operate PostgreSQL with a verified recovery path

Identify the authorized cluster/database, version, hosting model, topology and operating commitments. Reuse accepted recovery objectives and procedures. Read sources for version-specific PostgreSQL and reviewed upstream material. Managed-service controls differ from self-hosted server access; do not invent superuser, filesystem or control-plane capabilities.

Establish recoverability

Begin with what must be recovered: one record set, one database, or the cluster and its supporting state. Establish the acceptable recovery point and time, then choose a backup and restoration approach that can meet them in the actual hosting environment. Logical backup, physical backup/WAL recovery and managed snapshots/PITR serve different recovery units and operating constraints; a combination may be needed.

Check the complete restore path, including access to the backup, compatible tools, required roles/extensions and the configuration needed to start the recovered system. A backup that cannot be restored within the available access and time does not meet the operating requirement.

  • Include roles/global objects, extensions, encryption/access, dependencies and recovery configuration.
  • A replica shares many failure and deletion modes and is not an independent backup.
  • Continuous recovery depends on a usable base and the required WAL chain; monitor retention and restoreability, not just successful job completion.

Exercise recovery on an authorized isolated target. Verify a known recovery point, application compatibility, representative records/invariants and access, then measure actual recovery time and attainable data loss. Keep restored deletion/retention obligations and traffic cutover explicit.

EvidenceClaim supported
Documented procedureA recovery approach is described.
Successful backupThe backup operation completed.
Successful restoreThe restoration was exercised.

None automatically establishes full application recovery.

Maintain health without hiding risk

Assess connections, long transactions, disk/WAL growth, vacuum/freeze, replication slots and lag against workload and retention.

  • Slots can retain WAL; replicas can lag or conflict with recovery.
  • Investigate cause and consequences before dropping slots, cancelling work or changing durability.
  • Define primary/failover authority and fencing against split-brain; promotion alone does not prove clients moved or application writes are correct.

When upgrading

For minor/major upgrades, check supported server/extension/driver combinations, storage needs and tooling.

  • Compare in-place, logical replication or dump/restore strategies against downtime and reversibility.
  • Rehearse meaningful data and query/role compatibility, validate the new target, and define cutover, stale-writer prevention and forward recovery.
  • Avoid claiming binary rollback after new writes unless that path is demonstrated.

Execute and report within scope

Use the named target and supported operational tooling. Confirm impact, resource/traffic boundaries and recovery route before consequential live work; existing authorization remains valid. Continue independent planning when access is unavailable. Protect backups and reports from secrets/personal-data exposure, retire temporary restored targets only within authorized cleanup, and report actual actions, evidence, remaining risks and operating ownership.

All versions