dpla-verify-and-notify - SKILL.md Agent Skill

name: dpla-verify-and-notify description: Verify ingest outcomes and send failure or status notifications to Slack or tech@dp.la. Use when the user asks to verify the ingest, check if it succeeded, notify about a failure, or post to tech-alerts.

DPLA Verify and Notify

Purpose

After a run or when something fails, verify pipeline output and ensure the right people are notified (Slack #tech-alerts or email tech@dp.la).

When to Use

"Verify the ingest"
"Check if it succeeded"
"Notify about the failure"
"Post to tech-alerts"
"Send failure alert"
"Who do we notify when ingest fails?"

Environment: For commands that run the orchestrator or scripts, ensure source .env has been run in the shell so JAVA_HOME, SLACK_WEBHOOK, etc. are set.

Verification

Pipeline success markers

Each step writes _SUCCESS when complete:

# Harvest
ls $DPLA_DATA/<hub>/harvest/<timestamped-dir>/_SUCCESS

# Mapping
ls $DPLA_DATA/<hub>/mapping/<timestamped-dir>/_SUCCESS

# Enrichment
ls $DPLA_DATA/<hub>/enrichment/<timestamped-dir>/_SUCCESS

# JSONL
ls $DPLA_DATA/<hub>/jsonl/<timestamped-dir>/_SUCCESS

Record counts

cat $DPLA_DATA/<hub>/harvest/<timestamped-dir>/_MANIFEST
cat $DPLA_DATA/<hub>/mapping/<timestamped-dir>/_SUMMARY

Escalation reports (orchestrator runs)

When the orchestrator has failures, it writes:

data/escalations/failures-<run_id>.md — human-readable
data/escalations/failures-<run_id>.json — machine-readable

ls data/escalations/
cat data/escalations/failures-*.md | tail -1

What to Report (failures)

Stage	What to report
Harvest	Feed unreachable, timeout, or harvest step failed
Mapping/remap	Mapping or IngestRemap failed (non-zero exit, no output)
Sync	S3 sync failed or was blocked
Anomaly	Anomaly detection halted sync (critical threshold)

Include: hub name, stage that failed, one-line error or path to logs/escalation report.

Where to Notify

Slack #tech-alerts: Preferred when SLACK_WEBHOOK is set. Post a short message: hub, stage, error or link to report.
Email tech@dp.la: Use when Slack is unavailable or as backup. Same content: hub, stage, error or attach/link failure report.

When using the orchestrator, it posts failure alerts automatically; when running scripts manually, you (or the agent) must post or email.

Test notifications

To verify webhooks without running an ingest:

source .env
./venv/bin/python -m scheduler.orchestrator.main --dry-run-notify

Messages are sent with a [TEST] prefix.

Reference

Full notification policy: AGENTS.md (sections on notifications and #tech-alerts)