astra-admin

name	astra-admin
description	Inspect and clean up AstraDB collections. Use when the user asks to "list astra collections", "clean up astra", "delete a collection", "fix the 100 index limit", "free up astra space", or when an astra write fails with "Cannot have more than 100 indexes". AstraDB free tier caps each keyspace at 100 Storage Attached Indexes (~25 collections). This skill wraps scripts/astra_cleanup.py so the agent can diagnose and recover.
license	MIT
compatibility	deepagents-cli

AstraDB Admin Skill

Lets the agent inspect the AstraDB keyspace and drop unused collections to free up the 100-index quota. Requires ASTRA_DB_API_KEY and ASTRA_DB_ENDPOINT in the environment.

Why this exists

AstraDB free tier: 100 Storage Attached Indexes per keyspace. Each collection costs ~4 indexes — so ~25 collections before new ones fail with Cannot have more than 100 indexes. Old test collections, old knowledge-base runs, and experimental data accumulate. This skill drops the stale ones to free slots.

The log warning AstraDB 100-index limit hit for 'X' — switching to get_collection is already handled gracefully in deploy/astra_store.py (the collection still works). But if you want to create a new collection, you have to free slots first.

Commands

List every collection with doc counts

python scripts/astra_cleanup.py

Prints a numbered table: name, doc count, and the cap warning.

Drop one specific collection

python scripts/astra_cleanup.py --delete collection_name_here

Prompts for [y/N] confirmation before dropping.

Interactive purge — keep a whitelist, review the rest

python scripts/astra_cleanup.py --keep agent_tasks,conversation_history,mem0_*

Walks every collection NOT matching the keep patterns and asks [y/N/q] for each one. q stops the loop. Prefix match with trailing * (mem0_* matches mem0_main, mem0_test, etc.).

Add --yes to auto-confirm every candidate (dangerous — only use when you already ran without --yes and know the list is correct):

python scripts/astra_cleanup.py --keep agent_tasks,conversation_history,mem0_* --yes

Core collections (NEVER delete these)

• agent_tasks — task lifecycle records for the dashboard
• conversation_history — raw message log, used as fallback when the LangGraph checkpointer is empty after a Render restart
• Any mem0_* — Mem0 semantic memory
• Any active knowledge-base collections

When the agent should use this skill

1. User reports AstraDB errors ("100 index limit", "collection create failed")
2. User says "clean up astra" / "list my collections" / "delete X from astra"
3. An agent tool call itself fails with the 100-index message and the agent wants to free space before retrying

Safety

• Dropping a collection is irreversible. The script always prompts unless --yes is passed.
• Always run without --delete/--keep first to see the current state.
• Never drop agent_tasks or conversation_history — the dashboard and history-replay-on-restart depend on them.

Reference

• Implementation: scripts/astra_cleanup.py
• AstraDB client in the app: deploy/astra_store.py
• AstraDB Python SDK: astrapy docs