Audit Trail

The upfront binary captures every thinking record produced during /feature runs and writes structured events to a local JSONL queue.

How it works

A PostToolUse hook fires after every /feature skill invocation. The hook:

1. Parses the thinking record from the tool response
2. Writes a structured event to .upfront/audit.jsonl
3. Optionally flushes to a remote endpoint

Events are stored locally first (durable — survives network failures), then flushed to your observability stack.

CLI commands

upfront status                            # Queue depth, last event, config
upfront log                               # View audit events (last 50)
upfront log --feature checkout --phase 1  # Filter by feature/phase
upfront flush                             # Push queued events to remote
upfront purge                             # Delete events older than TTL

Remote integration

Configure .upfront/config.json or ~/.upfront/config.json:

{
  "endpoint": "https://your-langfuse.example.com/api/public/ingestion",
  "auth_header": "Bearer pk-lf-...",
  "ttl_days": 90,
  "project_name": "my-project"
}

Compatible with Langfuse, Arize Phoenix, Helicone, Portkey, or any endpoint that accepts JSON POST.

What managers see

Three metrics without reading a single spec:

1. Adoption — percentage of features that went through /feature
2. Depth — how many phases were completed (all 4 = thorough, 1-2 = bailed early)
3. Effectiveness — rework rate difference between spec’d and unspec’d features

The audit trail is 30-second triage: decide whether to read the spec or send it back.