← ClayCONTENT HISTORY
WHAT CHANGED
Update to Clay
Snapshot Nov 18, 2026 · 06:19 UTC · version 2.38.0
Collection source: downloaded plugin package.
First saved snapshot
No earlier snapshot is available to establish a change.
Save Clay and check future changes in Following.
Up to 3 plugins free · no card required.
Compare saved observations
Download comparison JSONFull technical diff · 0 changed fields
Full snapshot data
{
"description": "Clay workflow alerts — use when the user wants to create, inspect, edit, enable, disable, or delete a workflow alert rule, or subscribe themselves to its email notifications.",
"included_files": [],
"name": "workflows-alerts",
"skill_md_contents": "---\nname: workflows-alerts\ndescription: Clay workflow alerts — use when the user wants to create, inspect, edit, enable, disable, or delete a workflow alert rule, or subscribe themselves to its email notifications.\nallowed-tools: Read\n---\n\n# Workflow alerts\n\nBefore using this skill, read the `clay` entry skill if you haven’t already this session, and follow its access guidance.\n\nAlert rules watch a workflow's published runs and notify its subscribers. Commands live under\n`clay workflows alerts`; run `clay workflows alerts <command> --help` for exact JSON shapes and\nerror codes.\n\n| Need | Command |\n| -------------------------------------- | -------------------------------------------------------------------------------- |\n| Conditions this workspace can use | `clay workflows alerts metrics list` |\n| A workflow's rules | `clay workflows alerts list <workflowId>` |\n| Create a rule | `clay workflows alerts create <workflowId> --input <json\\|file\\|->` |\n| Edit, enable, or disable a rule | `clay workflows alerts update <workflowId> <ruleId> --input '{\"enabled\":false}'` |\n| Delete a rule | `clay workflows alerts delete <workflowId> <ruleId>` |\n| Repeat interval and delivery schedules | `clay workflows alerts settings get <workflowId>` |\n| Recipients | `clay workflows alerts destinations get <workflowId>` |\n| Subscribe yourself to alert emails | `clay workflows alerts subscribe <workflowId>` |\n\nIf the alert operations are unavailable, consult the `clay` entry skill for access guidance\nor point the user to the workflow's Alerts panel in the Clay app. Do not work around the missing operations through direct API calls.\n\n## Workflow\n\n1. **Resolve the workflow and rule.** Use the workflow from the selected context, a supplied ID,\n or `clay workflows list`. Read it with `clay workflows get <workflowId>` and keep its real URL.\n Resolve rule IDs from `clay workflows alerts list`; never guess a rule ID, and ask when several\n rules match a name.\n2. **Read current state before any change.** Fetch `metrics list`, `alerts list`,\n `settings get`, and `destinations get`. Offer only metrics `metrics list` returns, for new rules\n and metric changes alike. Do not infer availability from metric names you know, enum\n membership, or feature flags. If the metric the user wants is absent, report it unavailable\n rather than picking another.\n3. **Translate intent with the returned condition.**\n - `condition.kind: \"aggregate\"`: choose one of its `operators`; give `threshold` in its `unit`\n (whole-number percent 0–100 or a run count) and one of its `windowMinutes`.\n Set `sampleFloor` only when `usesSampleFloor` is true and the user asked; it defaults to 10.\n - `condition.kind: \"event\"`: submit the returned `metric` as is, with no threshold, window, or\n floor, and do not ask for them. Use `{\"name\":\"Notify on every failure\",\"metric\":\"run_error\"}`\n only when the user wants to hear about failures themselves (\"notify me when it fails\", \"alert\n me on every failure\"), and do not substitute a failure-count rule. \"Immediately\" is delivery\n timing, not a condition: for \"alert me immediately when the failure rate exceeds 20%\", create\n the aggregate rule that condition names and handle the timing as below.\n `clay workflows alerts create --help` describes when a `run_error` rule fires, repeats, and\n resolves.\n - Timing follows the delivery schedules, not the rule. Before creating a rule for \"immediately\"\n or saying when alerts arrive, read `emailDeliveryCadence` and `slackDeliveryCadence` from\n `settings get`. If either is `daily` or `weekly`, tell the user first that the channel's\n alerts arrive in a daily or weekly digest, not immediately, and that the schedule is changed\n in the Clay app; create the rule only if they still want it. Say \"immediately\" only for a\n channel whose cadence is `immediate`.\n - `gte` with a threshold of 0 is rejected because it would fire on every check. For \"any\n failure in the window\", use `gt` with 0, or `run_error` when the user wants each failing\n streak.\n - Ask only for intent that is genuinely missing. If the user asks for a condition the\n catalog doesn't offer (for example run duration or credits), say it isn't available.\n4. **Avoid duplicates.** Match aggregate rules on metric, operator, threshold and window, plus\n sample floor only when the metric's `usesSampleFloor` is true (other metrics ignore a stored\n floor); match event rules on metric. Reuse an equivalent enabled rule; offer to enable a\n disabled match. Create a second similar rule only when the user explicitly asks for one. A\n workflow holds at most 10 rules. Evaluation timing is managed by Clay: there is no command or\n field to change how often rules are checked, so say so if asked.\n5. **Make the change.** Submit only the fields that change; omit an unchanged `metric` from\n updates. Do not change the workflow graph, publish, or run it. Explain that changing a rule's\n definition or deleting it closes its open alert; renaming alone keeps it open. A rule's name\n does not follow its condition: when you change the condition of a rule whose name describes\n the old one (for example \"Run failure rate (>20% in the last 1 hour)\"), send a matching new\n name in the same update, and keep a name the user chose. A run error rule that fires again after\n resolving sends a fresh first notification. Before saying an open alert will repeat, read\n `renotifyMinutes` from `settings get`: a number is the repeat interval in minutes, and `null`\n means it notifies once and never repeats. Describe timing only as the delivery schedules\n above; do not promise a time beyond that.\n6. **Keep email subscription separate.** Subscribing is explicit, workflow-wide, and only for the\n current user. Creating a rule subscribes no one. When the user says \"alert me\", check\n `isCurrentUserSubscribed` from `destinations get`, not whether the recipient list is empty.\n If it is false, ask whether to subscribe them by email; subscribing again is safe. Every rule,\n including `run_error`, notifies on the workflow's email and Slack delivery schedules; those\n schedules are read-only here. Changing schedules or the repeat interval, choosing a Slack\n channel, adding other users, unsubscribing, test sends, and resolving alerts happen in the Clay\n app. Point the user there instead of claiming success.\n7. **Handle results.** On `approval_required`, use the host's approval flow with\n `details.actionApprovalRequestId`; never approve on the user's behalf, and never rerun the\n command after approval, because approving runs it. On a `not_found` that says alerts are not\n available, report that; do not publish or run the workflow or try to turn on a flag.\n8. **Confirm from the server.** After a success or an approved change, read back with\n `alerts list` and `destinations get`. If the rule was created but a requested subscription\n failed, report the saved rule and the failed email setup; do not roll back and do not say\n notifications are set up. Report the rule name, condition, enabled state, the delivery\n schedules and recipients you read, and `[Open workflow](<url>)`. Do not report evaluation\n timing or flag state, and do not say a notification was sent because a rule was saved.\n\nRules evaluate published workflow runs; creating a rule does not publish or run the workflow.\n"
}SHA-256 of public snapshot: e24a3a81c51e789becfb984fb58e7a36f498203348a2b98ce97e4239918d6731