tmcfarlane/skillforge/cloudflare-log-reader

Cloudflare AI Gateway Log Reader

Read execution traces from Cloudflare AI Gateway via the Cloudflare API and store them in local SQLite for skill extraction and analytics.

When to Use

Use this skill to poll gateway logs after LLM calls complete, feed the AutoResearch loop with real usage data, or build a local analytics cache.

Steps

1. Get a Cloudflare API token with AI Gateway: Read permission scope. Store it in the environment as CF_API_TOKEN (not in code).

1. Build the request URL:

`` GET https://api.cloudflare.com/client/v4/accounts/{account_id}/ai-gateway/gateways/{gateway_name}/logs `` Query parameters:

• status=success — only successful completions
• order_by=created_at&direction=asc — chronological order
• since={iso_timestamp} — incremental polling from last seen entry

1. Authenticate with Bearer token:

``typescript headers: { Authorization: Bearer ${CF_API_TOKEN} } ``

1. Parse the response — result array contains log entries with fields: id, provider, model, status, cached, prompt_tokens, completion_tokens, duration, request, response, created_at.

1. Deduplicate by `id` before inserting — the API may return overlapping entries on incremental polls.

1. Insert into local SQLite gateway_logs table for offline analysis.

1. Flush DB to disk after batch insert (required when using sql.js WASM).

1. Track `MAX(logged_at)` as the cursor for the next incremental poll.

Algorithm

lastSeen = SELECT MAX(logged_at) FROM gateway_logs
entries = GET /logs?since=lastSeen&status=success
for entry in entries:
  if entry.id not in gateway_logs:
    INSERT INTO gateway_logs(...)
persistDb()

Primitives

• http-fetch, sql-query, scheduling, data-transform

Tags

cloudflare, sqlite, typescript, observability