DevOps workflows

When a PagerDuty incident resolves, post a customer-facing status update via webhook, then auto-draft a postmortem in Confluence from the incident timeline and Slack thread.

When a sustained cache hit-rate drop is confirmed, an agent investigates the offending GitLab config change, drafts a root-cause analysis with a concrete fix.

When a webhook reports Dropbox usage crossing a quota threshold, prunes the oldest non-release artifacts until usage drops under target and pages on-call via PagerDuty if it can't.

After a war-room bridge exists, this workflow looks up the affected service's tier and pages the matching escalation group.

An agent reviews each 90-day-stale flag, reads its call sites in the repo, judges removal risk, and posts a Confluence removal plan plus a Linear issue routed to the right owner.

An agent assembles a richer shift-handoff by reasoning across PagerDuty incidents, Sentry error spikes, and open GitHub PRs.

On a Honeycomb burn-alert during a canary, an agent correlates the regression against the GitHub diff, recent error traces, and infra changes.

Intercepts a request to promote a staging deployment to production, runs a final synthetic Web Vitals and bundle check.

Sends preview screenshots to a vision model that judges layout, broken images, and overflow.

When a Datadog monitor on Cloudflare cache-hit ratio fires, enriches the alert with the current cache breakdown and the last deploy.

When an origin-load spike is detected, an agent gathers Cloudflare cache data, the suspect Vercel deploy diff, and Datadog signals.

Captures screenshots of key pages on each Vercel preview, diffs them against approved baselines.

Each week it finds Linear flag-removal tickets that are past their due date, builds a ranked owner-by-owner digest.

Monthly, tallies total flags, how many are past kill-date, average overdue age, and per-team breakdowns from your code and Statsig.

Detects flags that have been toggled on permanently and behave like settings rather than experiments.

Quarterly, an agent inventories every feature flag, scores each for staleness and risk, opens cleanup PRs for clear wins, files Linear tickets for the ambiguous ones.

On a Sentry error-spike alert, an agent reads the stack trace, checks for an existing duplicate issue.

Captures each Vercel preview across mobile, tablet, and desktop widths, overlays them against the matching Figma design frames.

Each week an agent reviews all open flaky-quarantine issues across GitHub and Linear, ranks the worst offenders by age and impact, writes a Notion review page.

When a preview deploy fails its performance budget, an agent investigates the diff to identify the likely cause, drafts a remediation plan.

On a critical-severity base-image CVE re-tag, opens a rebuild PR, pages the on-call engineer in PagerDuty.

Checks each tracked base image against its EOL date and, when a production image crosses into the final countdown window with no open rebuild PR, raises a PagerDuty incident…

When Sentry detects a new error spike, find the GitHub deploy that likely caused it, summarize the regression with an LLM, and post a candidate-cause report to Slack.

An agent reviews all upcoming base-image EOL deadlines across the org, weighs each by blast radius and remaining runway.

An agent crawls the repository on demand, reconciles in-code flag references against the flag registry.

On a weekly schedule, aggregates the current quarantine manifest and recent flake history, builds a prioritized report.

Finds Vercel preview deployments older than their TTL, pings the deploy owner in Slack with a grace window, then deletes any preview still stale after the window expires.

Listens for merged GitLab MRs labeled flag-removal, marks the flag retired in Postgres, and closes the matching Linear cleanup issue so the debt ledger stays accurate.

An agent reviews stale preview and staging environments across GitHub and Vercel, reasons about which are safe to reap versus needs-owner-input, asks ambiguous owners in Slack.

On every Cloudflare Worker deploy, compares post-deploy cold-start p95 against the previous deploy tag's baseline and blocks/flags the release in GitHub if latency regressed past…

Pulls the live container inventory from Datadog, maps each running image to its base layer's support status.

Lets an engineer request removal of a named flag from Slack; the agent verifies it is past sunset and carries no live traffic, then opens a removal PR or explains why it refused.

On a quarantined test, a CEO-driven agent reads the test source and recent diffs, diagnoses the likely cause of nondeterminism, opens a draft fix PR.

When a Vercel preview passes its smoke gate, posts a Slack message to the team channel summarizing the checks and tagging the PR author that the preview is verified and ready…

An agent reviews each long-rolled-out flag, reads its code references and recent commit history, drafts a removal plan.

Posts an interactive QA checklist to Slack when a Vercel preview is ready, collects an Approve or Block decision from an on-call reviewer.

Triggered by a flag-management webhook the moment a flag reaches full rollout, it waits out a soak period, confirms no error spike.

Checks Datadog flag-evaluation metrics weekly and proposes retiring any flag that has had zero evaluations for 30 days, opening a draft PR and a Linear cleanup ticket.

When a high-urgency PagerDuty incident is triggered, spin up a dedicated Slack channel, start a Zoom bridge, create a tracking Linear issue, and invite the on-call responders.

On an Axiom budget-burn alert webhook, enriches the alert with the offending dataset and growth delta.

After each GitHub release is published, downloads the matching build artifact from Dropbox, checks its integrity, and confirms the release is restorable.

Compiles a weekly inventory of every base image in use with its support status and days-to-EOL, writes the snapshot to BigQuery for trend tracking.

When a GitHub deployment status flips to failure, this workflow opens a Zoom incident bridge, gathers the commits since the last good deploy.

When a preview smoke gate fails, an agent pulls the build logs and failing responses, diagnoses the likely cause, files a Linear issue, and comments the analysis on the PR.

Before opening a rebuild PR for a re-tagged base image, scans both the old and new digests and only proposes the bump when the new image clears the CVEs that triggered the alert.

When a test is quarantined, an agent reads its recent failure logs, infers the likely root cause (timing, ordering, network, fixture), drafts a remediation plan.

When a production deploy breaches both the bundle-size and build-time budgets, pages the on-call engineer through PagerDuty and posts the candidate rollback target…

On a weekly schedule, pull Datadog alert volume and PagerDuty incident history, have an agent flag noisy monitors and overloaded rotations, and file tuning tasks in Linear.

Compiles a monthly feature-flag debt report from the registry and Datadog usage, publishes it to Confluence, and pings each team in Slack with their overdue-flag count.

On each flag-config change pushed to GitHub, it captures who changed what plus a Honeycomb health snapshot of the affected cohort and appends an immutable audit row to BigQuery…

Compiles every flag at 100% for 30+ days into a ranked debt report — count, age, and owning team — publishes it to Confluence and posts a summary to Slack.

When a preview smoke check fails, an agent inspects the browser trace and console errors, determines the likely root cause.

When a preview smoke check fails, an agent inspects the browser trace and console errors, determines the likely root cause.

Aggregates the week's flaky-test occurrences from GitHub and Datadog into a ranked health report, publishes it as a Confluence page.

Polls Honeycomb for the error rate on a newly rolled-out flag cohort and, if it breaches your threshold versus the control group, disables the flag via GitHub and pages…

On a Datadog service alert, the workflow checks GitHub for deploys in the last 30 minutes and, if a recent deploy lines up, pages on-call with a rollback recommendation instead…

When a Vercel deployment fails or its post-deploy health check breaks, page on-call in PagerDuty, open a GitHub revert PR for the bad commit, and notify the team in Slack.

When a GitLab MR merges it records an open lead-time entry in Postgres; a Honeycomb deploy marker later closes the matching entry with actual deploy time.

When an engineer runs a Slack command, this workflow creates a Zoom bridge, opens a tracking Sentry-linked incident, files a Linear issue for follow-up.

When Datadog detects an edge-function cost anomaly after a production deploy, an agent correlates the spike to the deploy, decides whether to roll back via the Vercel API.