clear-tinker-storage

name	clear-tinker-storage
description	Delete all checkpoints from your Thinking Machine Tinker prod account to reduce storage charges. Requires TINKER_API_KEY.

Clear Tinker Storage

Delete all checkpoints from the authenticated Tinker account to free up storage.

Instructions

When this skill is invoked, perform the following steps:

1. Load the API key from docker/.env.tinker (the TINKER_API_KEY variable). If not found, ask the user to provide one.

2. List all checkpoints by paginating through the Tinker REST API:

   GET https://tinker.thinkingmachines.dev/services/tinker-prod/api/v1/checkpoints
   Headers: X-API-Key: <key>, User-Agent: CLaaS/0.1
   Params: limit=100, offset=0
   

   The response JSON has checkpoints (array) and cursor (with total_count, offset, limit). Paginate by incrementing offset until all checkpoints are fetched. Important: A custom User-Agent header is required — Cloudflare blocks Python's default user-agent.

3. Display a summary to the user before deleting:

   • Total number of checkpoints
   • Total storage in GB
   • Breakdown by checkpoint_type (training vs sampler)

4. Ask for confirmation before proceeding with deletion. This is a destructive, irreversible operation.

5. Delete each checkpoint by parsing the tinker_path and calling:

   DELETE https://tinker.thinkingmachines.dev/services/tinker-prod/api/v1/training_runs/{training_run_id}/checkpoints/{checkpoint_id}
   Headers: X-API-Key: <key>
   

   The tinker_path format is tinker://{training_run_id}/{checkpoint_id}. Split on the first / after removing the tinker:// prefix to extract both parts.

6. Verify deletion by re-querying the list endpoint and confirming 0 checkpoints remain.

7. Report results: number deleted, number failed, storage freed.

Authentication

• The API key must be a valid Tinker API key (starts with tml-)
• Auth header format: X-API-Key: <key>
• If the key returns 401, ask the user for a fresh key

Notes

• All API calls use httpx with a 30-second timeout
• This deletes checkpoints across ALL training runs for the authenticated user
• There is no undo — deleted checkpoints cannot be recovered
• After clearing, new LoRAs must be re-initialized via POST /v1/lora/init