Datasette Skill | Agent Skills

Datasette

Datasette (datasette.io) is the CLI and runtime for publishing SQLite files as a read-only web application and JSON API. It sits on top of the same .db file format as the sqlite engine: this skill is the tooling and ecosystem layer (serve, metadata, plugins, CORS), not a second database engine.

sqlite → datasette: you pass the same path as sqlite3 my.db. Datasette does not build a second on-disk database for serving—read-only HTTP over that file. That is the usual sense of “zero copy” live publishing here: no duplicate .db artifact; refreshes see committed data from other writers when the file uses a normal WAL setup. For a static snapshot, serve a copy or use --immutable on a file you do not change.

What you get

datasette — local web server; table browse, filters, Custom SQL, export links.
Metadata YAML — per-database titles, facets, sortable columns, canned queries (named SQL with optional :parameters).
--crossdb — attach multiple SQLite files; run SQL that references more than one database (see Datasette docs for exact SQL shape).
Plugins — ecosystem extends CSV export, maps, full-text helpers, etc. (Plugin directory).
JSON API — append .json to many URLs for programmatic use (JSON API).

Official documentation: https://docs.datasette.io/

Install

Use pip / pipx / your OS package manager per current Datasette docs. You need the datasette command on PATH.

Basic serve

datasette mydata.db

Opens a local port (default in docs) with table list and row browsing.

Metadata

Point at a YAML file with -m / --metadata. It can define title, description, per-databases, per-tables (facets, labels), and queries (canned SQL surfaced in the UI).

cursor-mirror integration

MOOLLM’s cursor-mirror skill ships a consolidated export and a large metadata file tuned for introspection:

| Artifact | Role | |----------|------| | lib/datasette_export.py → export_datasette(output_path) | Builds a single cursor-mirror.db with tables such as composers, transcript_sections, shell_commands, tool_calls, usage_events, feature_flags (see module docstring). | | reference/universal/datasette-metadata.yml | Datasette metadata: canned queries (tool usage, shell audit, cost, timelines), facets, cross-database layout for cursor-mirror + cursor-model. | | reference/universal/TIME-INDEXED-TABLES.md | Notes on time columns for bucketing and correlating with other SQLite DBs in the same Datasette process. |

Typical flow

From the cursor-mirror skill directory (or with PYTHONPATH set so lib imports resolve), run Python to call export_datasette(Path("cursor-mirror.db")) (see datasette_export.py for parameters and defaults).
Ensure cursor-model.db exists if you want the universal model side (same metadata file names both databases).
Serve with cross-database and metadata:

datasette cursor-mirror.db cursor-model.db --crossdb \
  -m reference/universal/datasette-metadata.yml \
  --cors -p 8001

Adjust paths to your checkout. Extra SQLite files (e.g. other Cursor history DBs) can be additional positional arguments; they appear as separate databases and can be joined in Custom SQL when timestamps align (see TIME-INDEXED-TABLES.md).

Security: Datasette is powerful; do not expose it on untrusted networks without authentication and threat modeling. Treat exported DBs as sensitive (chat content, tool history, shell commands).

Relationship to other skills

| Skill | Role | |-------|------| | sqlite | Engine and .db file semantics; sqlite3 CLI for direct SQL. | | cursor-mirror | Source data and export; mirror CLI and YAML model docs. | | schema | Schemapedia datasette mechanism points here. |

Related skills

sqlite — file format, pragmas, JSON1 when you edit the DB outside Datasette.
cursor-mirror — full introspection protocol and mirror commands.

Part of MOOLLM

This skill's directory (browse and fetch everything): skills/datasette/

MOOLLM: repo · Skill index and docs: skills/README

Agent Skills: Datasette

Install this agent skill to your local

Skill Files