CocoIndex skill
The CocoIndex skill teaches an agent version 1 of the incremental pipeline library: apps, target states, memoization, connectors.
- 11,889 ★ on the repository
- Last commit
- Apache-2.0
Minimal access
Why this level
- The skill is plain instructions; pipelines the agent writes will reach databases and LLMs under your keys when run.
Install
- Works in
Manual install
Copy the `skills/cocoindex` folder of cocoindex-io/cocoindex into the agent skills directory, then install the library with `pip install -U cocoindex`.
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill is plain instructions; pipelines the agent writes will reach databases and LLMs under your keys when run.
This is third-party code. Review the repository files before installing.
What it does
The skill in `skills/cocoindex` describes CocoIndex, a Python library for pipelines that recompute only changed data. The principle is written as a formula: target state is computed from source state, and the library detects changes and syncs the result itself. The agent gets the basics (apps, function memoization, target states) and five references: API, connectors, patterns, database setup and project setup. The skill covers PDF and Markdown processing, vector embeddings, LLM extraction, knowledge graphs, ETL between databases and Kafka streams. Sync targets are PostgreSQL, SQLite, LanceDB, Qdrant, SurrealDB, Apache Doris, the file system and Kafka.
Who it is for. Python developers building indexes, RAG and ETL on CocoIndex who want the agent to write version 1.0 code without outdated APIs.
Good fit when
- A pipeline must embed documents and refresh only changed files.
- The agent writes a first CocoIndex project with `cocoindex init`.
- Data moves from one database to another with automatic change detection.
- Entities and relations are extracted from text into a knowledge graph.
Not a fit when
- The project runs CocoIndex before version 1.0: the skill covers only the new API.
- A ready code-search MCP server is required: the separate cocoindex-code project covers it.
- Data is processed by a one-off script with no change tracking.
Example request
Create a CocoIndex project that reads Markdown from the docs folder, chunks it, computes embeddings and writes them to PostgreSQL.Limitations
The skill targets CocoIndex 1.0.0 and later. The `.agents/skills` and `.claude/skills` folders in the repository serve development of the library itself (releases, backports, diagrams) and are not for users. Steps to install the skill into an agent live in the CocoIndex documentation, not the README. Embeddings and LLM extraction need your own models or keys. The skill teaches the Python API and does not replace decisions on the database, chunk size and embedding model; name them in your request to the agent.
How to disable. Delete the `cocoindex` folder from the agent skills directory. `pip uninstall cocoindex` removes the library itself.
FAQ
Which CocoIndex version does the skill cover?
Version 1.0.0 and later. Code from the skill will not fit older projects.
How do I run the pipeline the agent wrote?
With `uv run cocoindex update main.py`, which installs dependencies and runs the app. `cocoindex init my-project` creates a new project.
Where can the skill write results?
To PostgreSQL, SQLite, LanceDB, Qdrant, SurrealDB, Apache Doris, files and Kafka.
How does the skill differ from cocoindex-code?
The skill teaches an agent to write pipelines on the library. The README describes cocoindex-code as a separate MCP server with a code index for Claude Code and Cursor.
Does CocoIndex itself need a separate database?
No. The generated `main.py` sets a local `cocoindex.db` file for internal data. The target database is chosen by task from the supported list.
Discussion
No questions or reviews yet.
Related
SkillsAround 200 skills for Claude Code on bioinformatics, chemistry and biostatistics, from RNA-seq analysis to molecular docking
Nine Pinecone skills: the agent creates a vector index, loads data, searches it and builds an assistant for documents
Voxel51 skills for computer vision: import datasets, find duplicates, run inference and evaluate models in FiftyOne
Probabl skills for machine learning experiments: a scikit-learn and skrub pipeline, evaluation, audit and a link to Skore Hub
17 skills for computational biology: RNA-seq, single-cell, protein structure, docking, literature and figures
31 skills for Claude Code and Codex for the analyst workflow: data checks, cohorts, funnels, A/B tests, dashboards, summaries
README in short
The README points to the skill with a line promising correct version 1 code from the agent and sends readers to the docs for install steps.