Replit Monitoring Agent

Investigate alert [identifier] for [app] without editing code or settings.
Build a timeline from deploys, logs, metrics and health checks. Identify the failing user journey, reproduction steps, likely cause, confidence and alternative hypotheses.
Redact sensitive values. Return the smallest next diagnostic and stop; do not fix, republish, restart services or change monitoring.

For reproduced incident [identifier], implement the smallest fix on a reviewable change set.
Add a regression test, run [required commands], compare relevant metrics in a safe environment and document rollback. Preserve public behavior outside [scope].
Do not republish, migrate production data, rotate secrets or change alert thresholds. If validation is incomplete, return a diagnostic patch rather than a claimed fix.

Investigate [integration] failures during [time window].
Separate our code, credential/permission, provider outage, quota/rate-limit and malformed-data hypotheses. Use official provider status and documentation plus redacted local logs.
Do not print or rotate secrets, retry writes repeatedly or disable verification. Recommend a safe fallback and escalation threshold.

Assess incident [identifier] for possible database involvement using read-only queries and schema inspection.
Identify affected tables, query paths, migration history, lock or latency signals and potential data impact. Produce a backup and rollback checklist.
Do not mutate data, run migrations, change indexes or expose row contents. Escalate immediately if integrity loss is plausible.

After deployment [id], validate [critical journeys] for [time window].
Run [smoke checks], compare errors and latency to [baseline], and inspect logs for new signatures. Return pass, investigate or rollback-recommended with evidence.
Do not auto-republish or roll back. Stop at [time cap] and mark missing observability as unknown.

Every day at [time], summarize the previous 24 hours for [app].
Report confirmed incidents, alert noise, deploys, error and latency changes, unresolved risks and recommended work. Link each item to logs, metrics or a change.
Do not claim uptime percentages unless the measurement covers the full interval. Do not edit code or settings from this report.

Compare [current window] with [baseline window] for route, job or query [scope].
Control for traffic volume and deploy changes. Identify the slow path with profiles or traces where available and propose one reversible optimization plus a benchmark.
Do not trade correctness, security or cost for a headline speedup. Label missing instrumentation and avoid causal claims without evidence.

Draft a blameless incident review for [identifier] from approved logs, changes and decisions.
Include impact, detection, timeline, root cause, contributing factors, recovery, what went well, what failed and action items with owner suggestions and verification criteria.
Do not invent user counts, duration or causality. Mark every missing fact and keep personnel speculation out.