Cursor Automations Setup
Cursor Automations are the cloud runner for wiki jobs that can operate from a fresh repo clone plus MCP tools.
Routing
Open this page when creating, auditing, repairing, or explaining Cursor cloud automations. Use Operational Heartbeat and npx tsx scripts/check-freshness.ts for run freshness; use this page for setup shape, cloud constraints, and which jobs should stay local.
As of 2026-11-03, the repo has 11 canonical automation definitions under automations/; 2 have cloud-adapted .cursor-automation.md prompts. Local-only jobs need filesystem credentials, browser sessions, local transcripts, or cost-gated manual review. Fourteen definitions, twelve with cloud prompts, were retired on 2026-11-03; their workflows run on request. Source: find automations, 2026-11-03
Cloud-Ready Automations
These have <slug>.cursor-automation.md files and can be configured at cursor.com/automations:
| Slug | Schedule | Primary MCP / source |
|---|---|---|
self-heal |
on-demand | GitHub |
source-compile |
every 4h locally; daily-compatible in cloud | repo-committed raw sources |
Local or Manual Automations
| Slug | Why not normal cloud |
|---|---|
calendar-sync |
Google credentials / local integration setup |
email-sync |
Gmail credentials / local integration setup |
slack-sync |
Slack browser session / local agent-browser setup |
meeting-sync |
Circleback token / local integration setup |
cross-project-learning |
local transcripts across projects |
security-scan |
cost-gated, target-repo-specific, manual review |
Setup Flow
For each cloud-ready automation:
- Open cursor.com/automations.
- Create a new automation.
- Set repository to
kevin-wiki. - Set the schedule from the automation frontmatter.
- Enable required MCPs: Firecrawl for web scans, GitHub for repo data.
- Paste the instructions from
automations/<slug>.cursor-automation.md. - Run once manually if the UI supports it.
- Confirm a PR or output file appears.
- Merge the PR so
state.jsonupdates freshness.
Cloud Runner Constraints
Cursor cloud agents run from a fresh sandbox. They cannot read:
~/.cursor/projects/transcripts~/.codex/sessions/- local browser profiles
- local OAuth tokens
- uncommitted files on Kevin's machine
If an automation needs those, run it locally or commit the raw source first.
Output Contract
Cloud automations should:
- read
automations/_cursor-automation-preamble.md - write
outputs/<YYYY-MM-DD>/<slug>/cursor-automation.md - update wiki targets only when evidence supports it
- update
state.json - append
wiki/log.md - run generated-surface closeout commands from Generated Surface Contract when generated packs, registries, redirects, qmd, or stateful ledgers changed
- open a PR
Monitoring
Use:
npm run audit:automations
npx tsx scripts/check-freshness.ts
The readiness audit is authoritative for installed/configured prerequisites and live Codex schedule drift. The freshness script is authoritative for overdue and never-run status. Cursor's web UI is the source for cloud run history and cloud MCP enablement.
Timeline
- 2026-11-03 | Retired fourteen legacy definitions and their twelve cloud prompts; refreshed counts to 11 canonical automations / 2 cloud prompts. Source: Kevin approval, 2026-11-03
- 2026-07-14 | Removed the obsolete Hawaii trip and offsites planning definitions and live Codex schedules; refreshed counts to 22 canonical automations / 14 cloud prompts. private page
- 2026-07-01 | Added automation proof and generated-surface closeout to the Cursor cloud output contract. private page
- 2026-06-30 | Updated counts to 24 canonical automations / 16 cloud prompts and added Slack sync as a local-only browser-session job. private page
- 2026-06-18 | Updated stale automation counts and replaced the old 8-of-9 table with current cloud-ready vs local/manual automation setup. private page
- 2026-04-07 | Initial setup guide for cloud-hosted Cursor Automations. Source: automations/_schema.md