agentrium: Command for Claude Code

Retrieve end-user error reports from ct-analytics

You are pulling end-user error/exception reports stored by the ct-analytics Cloudflare Worker (D1 table errors) and writing a human-readable summary inline. Errors are populated by POST /error_report from three sources: rust_panic, rust_command, frontend. Retention is 90 days.

Arguments

$ARGUMENTS - Window in days (e.g. 1, 7, 30). Plain integer or with trailing d (e.g. 7d). Defaults to 7 if empty. Clamped to [1, 90].

Step 1: Resolve the window

1. Parse $ARGUMENTS. Strip a trailing d if present. Coerce to integer. If empty or invalid, use 7.
2. Clamp to [1, 90].
3. Tell the user the resolved window (e.g. "Pulling errors for the last 7 days...").

Step 2: Try the Worker endpoint first (preferred)

1. Resolve the stats token, in this order:
   • $env:CT_STATS_TOKEN if set
   • Otherwise, retrieve STATS_TOKEN from your auto-memory at ct_analytics_tokens.md (memory type reference, slug ct-analytics-tokens). Do not print the token to the user or include it in any saved file.
2. If a token is available, call:

   GET https://ct-analytics.claude-terminal.workers.dev/errors/summary?days={N}&limit=20
   Header: x-ct-token: {STATS_TOKEN}
   

   On Windows PowerShell, use Invoke-RestMethod with -Headers @{ 'x-ct-token' = $token } and capture to a variable; do not echo the token.
3. If the request returns HTTP 200 with JSON, parse it and skip to Step 4.
4. If the request returns 404 (endpoint not deployed yet), tell the user the Worker needs a one-time deploy and offer to run it:

   Run from workers/ct-analytics/: npx wrangler deploy
   

   Ask before running. If they decline, fall through to Step 3.
5. If the request fails for any other reason (no token, non-2xx, network error), fall through to Step 3.

Step 3: Fallback - query D1 directly via wrangler

Run wrangler from workers/ct-analytics/ against the production D1 (--remote). Use --json so the output is parseable. Run all queries; capture stdout per query.