Agent Skills › topoteretes/cognee › cognee-migrations

cognee-migrations

GitHub

用于处理 Cognee 数据库迁移,包括理解自动执行机制、使用 CLI 检查或修复迁移状态、编写 Alembic 修订版及跨系统数据迁移。

.agents/skills/cognee-migrations/SKILL.md topoteretes/cognee

触发场景

处理数据库迁移相关任务 使用 cognee-cli 检查或修复迁移状态 编写新的迁移脚本

安装

npx skills add topoteretes/cognee --skill cognee-migrations -g -y
更多选项

非标准路径

npx skills add https://github.com/topoteretes/cognee/tree/main/.agents/skills/cognee-migrations -g -y

不安装直接使用

npx skills use topoteretes/cognee@cognee-migrations

指定 Agent (Claude Code)

npx skills add topoteretes/cognee --skill cognee-migrations -a claude-code -g -y

安装 repo 全部 skill

npx skills add topoteretes/cognee --all -g -y

预览 repo 内 skill

npx skills add topoteretes/cognee --list

SKILL.md

Frontmatter
{
    "name": "cognee-migrations",
    "description": "Use when dealing with cognee database migrations — understanding when they run automatically, checking or repairing migration state with cognee-cli upgrade\/downgrade\/stamp\/current, a write blocked by a failed migration, authoring a new Alembic (relational schema) revision or a graph\/vector data migration, or moving data between systems (relational DB import, memory export\/import)."
}

Database migrations

cognee has two migration chains, run together:

Chain Changes Lives in Revision stored in
Relational schema (Alembic) Tables and columns of the relational DB (users, datasets, ACLs, pipeline runs, …) cognee/alembic/ (alembic.ini is in cognee/) alembic_version table
Graph/vector data Cross-store data rewrites (re-keying node ids, adding graph columns) across graph DB, vector DB and relational ledger cognee/modules/migrations/ (registry.py, versions/) Per dataset: dataset_database.migration_revision (access control on). Globally: global_database_version.global_migration_revision (access control off)

Use it

They run by themselves

run_migrations() applies the relational chain first, then the data chain. It runs:

  • at API server startup (and in the Docker entrypoint.sh before gunicorn);
  • on the first write in an SDK or CLI process (remember, add, cognify, improve, memory imports), once per process;
  • when you call await cognee.run_migrations().

A fresh database is built by running the whole chain (no stamping). At head, the first run in each process still does a no-op Alembic upgrade plus a scan of the per-database revision rows; later calls in the same process are skipped by an in-memory flag. ENABLE_AUTO_MIGRATIONS=false turns off all automatic runs; then run cognee-cli upgrade yourself.

Concurrent processes are serialized by a migration lock: a Postgres advisory lock (works across hosts) or a file lock next to the SQLite DB (one host only).

Check and repair

cognee-cli current                   # stamped revision per database, and the last failure
cognee-cli history                   # the data chain, newest first
cognee-cli upgrade                   # relational to head, then data chain to head
cognee-cli upgrade <slug>            # data chain up to and including <slug>
cognee-cli upgrade --alembic <rev>   # pin the relational target
cognee-cli downgrade <slug|base> [--dataset UUID ...] [--alembic REV] [--force]
cognee-cli stamp <head|base|slug> [--dataset UUID ...] [--force]
  • The positional revision is always a data-chain slug; relational targets go through --alembic.
  • downgrade rewrites data and asks for confirmation. It only reverts spans where every migration defines a down(), and leaves the relational schema alone unless you pass --alembic.
  • stamp changes only the stored data-chain revision, without running anything. Use stamp base --dataset <id> and then upgrade when a database's data drifted from its stamp (for example after restoring a backup); the chain is idempotent and converges it.
  • upgrade runs even with ENABLE_AUTO_MIGRATIONS=false.

A write is blocked

If a dataset's data migration failed, writes to that dataset are refused until it succeeds (with access control off, any failure blocks all writes). The server still starts. Run cognee-cli current to see the error, fix the cause, then cognee-cli upgrade. A failed run is retried on the next start or write.

Moving data between systems (not schema migrations)

Goal Use
Turn an existing relational database into a graph migrate_relational_database(graph_db, schema) (cognee/tasks/ingestion/migrate_relational_database.py), with the source DB set by MIGRATION_DB_PROVIDER / _PATH / _NAME / _HOST / _PORT / _USERNAME / _PASSWORD. Examples: examples/demos/ingestion_and_migration/
Back up a dataset, or move it to another cognee instance A COGX archive, see below
Export a dataset's graph for other tools await cognee.export(dataset, format=...): "json", "graphml" or "cypher" write a file (one way: cognee can't import them back); "pydantic" (default) returns typed DataPoint objects in memory
Import from another memory system (Mem0, Zep/Graphiti, Letta, LangMem) Build a MemorySource (cognee/modules/migration/sources/) and pass it to await cognee.remember(source, dataset_name=...)

There is no tool that moves a whole deployment from one database backend to another.

COGX archives

COGX (Cognee eXchange, cognee/modules/migration/cogx.py) is cognee's portable memory format and the only export format cognee can import back. An archive is a directory with a manifest.json (COGX version, source system, the dataset's data-migration revision) and one JSONL file per record kind (documents, episodes, entities, facts, memories, memory_blocks), plus nodes.jsonl with the raw graph nodes. The Mem0, Zep, Letta and LangMem importers also translate into COGX records first.

Since COGX 0.2 every record also carries an optional OMP header, the fields proposed to the Open Memory Protocol exchange group: origin (first system and id, kept across copies), content_hash (sha256 of the body, for dedupe; distinct from an integrity checksum), subject, semantic_kind (fact, decision, commitment, procedure, preference, observation, other), valid_from/valid_to with invalidated_by, evidence[] (role, ref, confidence), confidence (outcome, verified_at, verified_by, score), version, content_type (the body's format), tags, a record checksum stamped by the writer, and provenance as contributors[] (role, actor, kind, at) with the accountable human and ai_involvement. The cognee exporter fills what it can read from the graph and declares the rest in the manifest's coverage map; the loader keeps the header in each record's external metadata and on fact edges, so a re-export preserves it. 0.1 archives read unchanged.

Use it to back up and restore a dataset, or to copy one to another cognee instance:

from cognee.migration import COGXArchiveSource

await cognee.export("my_dataset", format="cogx", destination="backup_cogx")
await cognee.remember(COGXArchiveSource("backup_cogx"), dataset_name="my_dataset")
  • A restore defaults to mode="preserve": the archived graph is written back as-is, with no LLM calls. mode="hybrid" also re-cognifies the raw content; mode="re-derive" ignores the archived graph and extracts again (costs LLM tokens).
  • cognee.push() / cognee-cli push does the same to Cognee Cloud: it exports to COGX, packs it as a .cogx.tar.gz and uploads it, and the receiving instance restores it (preserve mode unless you pass mode=).
  • export(..., include_permissions=True) also writes permissions.json with the dataset owner and ACL grants, including password hashes, so the restore recreates working accounts. Treat that archive as a secret.
  • An archive written by a newer major COGX version is rejected with a ValueError; upgrade cognee on the importing side.

Pitfalls

  • Never regenerate cognee/alembic/frozen_schema.py. It is the certified base schema the initial revision builds from, pinned by cognee/tests/unit/test_frozen_schema_seal.py. Schema changes ship as new revisions at head.
  • Never hand-type an Alembic revision id. Hand-typed patterns (a1b2c3d4e5f6, …) already collided with a downstream chain that vendors this one (b2c3d4e5f6a7). Generate ids with alembic revision.
  • Never rename, remove, or reorder a data-chain entry. The slug is what deployed databases store; an unknown stored slug disables the chain for that database.
  • A model change and its migration land together. CI's "Migration/Model Lockstep Guard" fails otherwise.
  • The relational schema cannot be downgraded below the data-bookkeeping revisions unless the data chain goes to base in the same call.
  • alembic.ini is always the packaged one; COGNEE_ALEMBIC_PATH or --alembic-path only changes the scripts directory (for vendored chains).

How it works

run_migrations() (cognee/modules/migrations/startup.py) takes the migration lock, decides fresh vs existing (a users or alembic_version table exists), runs Alembic in-process on a worker thread, then walks the data chain per database with runner.run_database_migrations, stamping after every step. After the chain, it syncs vector-adapter storage when the recorded cognee_version differs from the library's (versions/adapter_storage_migration.py, not a chain entry).

  • Relational: cognee/alembic.ini, cognee/alembic/env.py, cognee/alembic/versions/, cognee/alembic/frozen_schema.py
  • Data chain: cognee/modules/migrations/ (README.md is the authoring contract; registry.py, migration.py, runner.py, startup.py, versions/)
  • CLI: cognee/cli/commands/migrate_command.py

Extending it

A new Alembic revision

cd cognee                              # the directory with alembic.ini
uv run alembic revision -m "add foo to data"

The DB URL comes from the live relational engine, so your normal .env settings apply. Follow the recent revisions (for example versions/e7f9a1c3d5b8_add_data_dataset_created_index.py):

  • Idempotent and guarded: inspect first (sa.inspect(op.get_bind())) and skip when the table is missing or the column/index already exists.
  • Branch on dialect (conn.dialect.name == "postgresql") for Postgres-only SQL. For Postgres enums use postgresql.ENUM(..., create_type=False) and create the type up front with checkfirst.
  • SQLite cannot drop or alter columns in place: use op.batch_alter_table(...).
  • Indexes: plain CREATE INDEX IF NOT EXISTS inside the migration transaction, not CONCURRENTLY (which releases the version-row lock and lets concurrent workers into the same build). Repair invalid Postgres indexes via pg_index.indisvalid.
  • A new model module outside the usual import path must be imported in cognee/alembic/env.py, or autogenerate will not see it.

A new data migration

Read cognee/modules/migrations/README.md first. In short:

  1. Write a module in cognee/modules/migrations/versions/ with async def migrate(context) (and optionally async def downgrade(context)); step 2 registers them as up= / down=.
  2. Append Migration(slug=..., cognee_version=..., up=..., down_revision=<previous slug>, down=...) to MIGRATIONS in registry.py. The chain is validated at import (linear, unique slugs).
  3. Make it idempotent and cheap on empty stores, and crash-safe: re-key derived stores (vectors, ledger) first and rename in the graph last.
  4. Freeze private copies of any logic you depend on; never import live models that may change later.

Tests

  • cognee/tests/e2e/migrations/test_migration_model_lockstep.py (+ schema_baseline.json): the lockstep CI job.
  • cognee/tests/unit/test_run_migrations.py: single Alembic head, startup behaviour.
  • cognee/tests/unit/test_frozen_schema_seal.py: frozen schema fingerprint.
  • cognee/tests/unit/modules/migrations/: data-chain unit tests.
  • cognee/tests/migrations/test_migration_lockstep.py: data chain against real stores (seed, down, up, verify).

版本历史

  • 525b1d1 当前 2026-11-02 22:48
  • 37f788b 2026-10-19 05:49

同 Skill 集合

.agents/skills/cognee-cli/SKILL.md
.agents/skills/cognee-community/SKILL.md
.agents/skills/cognee-custom-graph-models/SKILL.md
.agents/skills/cognee-custom-pipelines/SKILL.md
.agents/skills/cognee-docker/SKILL.md
.agents/skills/cognee-forget/SKILL.md
.agents/skills/cognee-improve-sessions/SKILL.md
.agents/skills/cognee-ingestion/SKILL.md
.agents/skills/cognee-integrations/SKILL.md
.agents/skills/cognee-performance/SKILL.md
.agents/skills/cognee-permissions/SKILL.md
.agents/skills/cognee-recall/SKILL.md
.agents/skills/cognee-server/SKILL.md
.agents/skills/company-brain-company-qa/SKILL.md
.agents/skills/company-brain-follow-up-agent/SKILL.md
.agents/skills/google-workspace-memory/SKILL.md
.agents/skills/multiagent-memory-setup/SKILL.md
.agents/skills/personalized-email/SKILL.md
.agents/skills/self-hosted-companion/SKILL.md
.claude/skills/cognee-cli/SKILL.md
.claude/skills/cognee-community/SKILL.md
.claude/skills/cognee-docker/SKILL.md
.claude/skills/cognee-integrations/SKILL.md
.claude/skills/cognee-permissions/SKILL.md
.claude/skills/cognee-server/SKILL.md
examples/demos/skill_feedback_loop/skills/skill-feedback-writer/SKILL.md
examples/demos/skill_feedback_loop/skills/pr-comment-evaluator/SKILL.md
examples/demos/skill_feedback_loop/skills/diff-risk-explainer/SKILL.md

元信息

文件数
0
版本
525b1d1
Hash
4639bcea
收录时间
2026-10-19 05:49

首页 - Wiki
Copyright © 2011-2026 iteam. Current version is 2.156.0. UTC+08:00, 2026-11-18 08:17
浙ICP备14020137号-1