| name | scheduled-workflow |
| version | 1.0.0 |
| description | Turns any GTM skill in this pack into a recurring Claude schedule — weekly prospect
refreshes, nightly CRM enrichment, daily pre-meeting briefings, monthly account
reviews, quarterly competitive scans. Invokes Claude's native scheduling feature
(the /schedule skill) under the hood, picks sensible cadences per workflow type,
enforces monthly credit ceilings so a runaway cron doesn't drain the balance, and
writes a plain-English schedule summary the user can edit later. Proactively offer
when a user finishes any workflow that would obviously benefit from running again
on a cadence.
|
| benefits-from | ["richapi-gtm","build-prospect-list","enrich-and-verify","account-research","pre-meeting-briefing","list-hygiene","competitive-intel","evidence-score"] |
| allowed-tools | ["Bash","Read","Write","Edit","Skill","AskUserQuestion"] |
| triggers | ["schedule this","run this every","automate this","set up a cron","nightly prospect refresh","weekly competitor scan","daily briefing","recurring enrichment","do this automatically"] |
scheduled-workflow
The "set it and forget it" specialist. Wraps Claude's scheduling feature around any GTM workflow in this pack so the user's go-to-market runs while they sleep.
Preamble
~/.claude/skills/richapi-gtm-skills/bin/richapi-skills-preflight
Phase 0 — is scheduling the right answer?
Not every workflow should repeat. Ask yourself (or the user):
| Signal | Schedule it |
|---|
| The data goes stale on a clock (titles change, companies hire) | ✅ yes |
| The user asks "just do this again next week / every Monday" | ✅ yes |
| The workflow produces a report a human will read | ✅ yes, scheduled delivery |
| The workflow touches 1000+ records | ⚠️ schedule but split across nights |
| The user needs the result right now, once | ❌ run it once, don't schedule |
| Every run spends > 500 credits | ⚠️ confirm the monthly ceiling first |
If any column-1 signal fires, offer scheduling.
Phase 1 — pick the cadence
Default cadences per workflow. The user can override.
| Workflow | Default cadence | Why |
|---|
build-prospect-list (top-of-funnel refresh) | Weekly, Mon 05:00 local | New leaders change jobs constantly; weekly catches the turnover |
enrich-and-verify (new CRM contacts since last run) | Nightly, 02:00 local | Keep CRM fresh without blocking daytime usage |
account-research (tier-1 named accounts) | Monthly, 1st of month | Firmographics + decision makers don't shift weekly |
account-research (target-account deep dive) | Quarterly | Heavy spend; quarterly captures meaningful change |
pre-meeting-briefing (calendar-driven) | Daily, 06:00 local | Brief runs before the user's morning; one-shot per meeting |
list-hygiene (pre-campaign checkpoint) | Weekly, Fri 16:00 local | Clean Friday → campaign Monday |
competitive-intel (ad + post sweep) | Weekly, Mon 07:00 local | Competitor ad creatives churn weekly; anything less misses rotation |
evidence-score (scoring backlog) | Nightly | Scores compound with new signal; run after enrichment |
check_usage (budget watchdog) | Daily, 09:00 local | Catch runaway spend within one day, not one month |
Phase 2 — set the cost ceiling
Every schedule must have a monthly credit ceiling so a regression or filter change doesn't drain the account. Ask:
"What's the most credits you want this schedule to spend per month? (default: 2× one-run cost × expected runs/month)"
The user's AskUserQuestion answer becomes the hard stop. If a scheduled run's pre-estimate exceeds the remaining monthly budget, the run skips and writes a note instead of spending.
Phase 3 — compose the schedule
Invoke Claude's native scheduling feature (the /schedule skill) with a complete brief:
Scheduled workflow — <workflow-name>
Cadence: <cron expression or human description>
Cost ceiling: <N credits / month>
On-run prompt:
Load the richapi-gtm skill. Run <sub-skill> with these filters: <JSON>.
If running cost would exceed <ceiling remaining>, skip and log.
On completion, <delivery method below>.
Delivery: email | Slack webhook | Claude conversation | write to file
Hand off to the /schedule skill to create the trigger. That skill knows the platform-specific cron semantics.
Cadence format cheat sheet
"every Monday at 05:00 local" — most readable; prefer this"0 5 * * 1" — standard 5-field cron"every 14 days" — for bi-weekly runs that don't land on a fixed weekday
Avoid sub-hour cadences. Nothing in this pack benefits from every-15-min polling — stale data isn't fresh data.
Phase 4 — choose a delivery method
Where does the result go when the user isn't there to see it?
| Delivery | When to use |
|---|
| Email to user | Default. Works everywhere. Best for reports. |
| Slack webhook | Team visibility. User pastes a webhook URL; we include it in the prompt. |
| CRM push via crm-sync-expert patterns | Enrichment schedules that update HubSpot/Salesforce/Pipedrive/GHL records directly. |
| Write to file | For developers running this on their own machine with Claude Code + cron. |
| Claude conversation | When the user will return to Claude to read results; scheduled triggers land in the next conversation. |
If the user doesn't specify, default to email and tell them it's changeable later.
Phase 5 — compose the run-time prompt
The scheduled trigger fires a prompt at Claude. Write it cleanly enough that the next session can execute it with no extra context. Template:
Run the weekly prospect refresh.
Skill: build-prospect-list
ICP: <frozen filter JSON>
Page limit: 3 pages (75 results max)
Cost ceiling: <N> credits — check check_usage first; abort if balance < ceiling.
Dedupe: suppress any LinkedIn URL already in our CRM.
Delivery: Email the new leads (not already-seen) to <user@example.com> as CSV.
If anything fails, leave a note in the conversation — don't retry.
Keys: it's self-contained, it names the skill, it carries the filters, it has a budget, it names the delivery method, and it has a fail-safe clause.
Phase 6 — write a plain-English schedule summary
After creating the trigger, produce a human-readable record the user can return to:
Scheduled: Weekly Prospect Refresh
What: build-prospect-list with fixed ICP
Cadence: Mon 05:00 America/Los_Angeles
Budget: 500 credits / month (alert at 80%)
Delivery: Email to you@yourdomain.com as CSV
Created: 2026-04-17
Edit / cancel: ask Claude to "update my weekly prospect refresh"
Offer to save this as a markdown file so they have a local copy.
Phase 7 — tell the user what the runtime contract is
Claude scheduling needs to be on. If the user's environment doesn't have it:
- Claude Code → works natively via the
/schedule skill
- Claude Desktop → triggers land in the next conversation; user must open Claude for the schedule to "catch up"
- Cursor / Windsurf / other MCP hosts → schedules are a Claude Code / Desktop feature; offer to set up a plain cron +
claude -p one-liner instead
Common scheduling recipes
Weekly ICP refresh → email to user
Cadence: Mon 05:00
Workflow: build-prospect-list
Fixed inputs: { seniority: ["Director","VP","CXO"], industries: ["SaaS"], geoIds: ["SF Bay Area"] }
Dedupe: against last run's output + user's CRM export
Budget: 500 credits / mo
Delivery: email CSV
Daily pre-meeting briefings from calendar
Cadence: Daily 06:00
Workflow: pre-meeting-briefing for every meeting that day
Inputs: pull LinkedIn URL from calendar event description or invitee email
Budget: 100 credits / week
Delivery: email one-per-meeting, or Slack to #morning-briefing
Monthly tier-1 account refresh
Cadence: 1st of month, 08:00
Workflow: account-research (light scan mode) on named-account list
Inputs: static list of 50 target accounts
Budget: 1500 credits / mo
Delivery: HubSpot company-record update via crm-sync-expert patterns
Daily usage watchdog
Cadence: Daily 09:00
Workflow: check_usage (free); alert if daily spend > 500 credits
Inputs: none
Budget: 0 credits (the tool is free)
Delivery: Slack to #ops only if threshold breached
Quarterly full competitive scan
Cadence: 1st of Jan/Apr/Jul/Oct, 07:00
Workflow: competitive-intel (deep-dive mode) across 5 named competitors
Budget: 2000 credits / quarter
Delivery: write report to /gtm/competitive/YYYY-QN.md + email summary
Anti-patterns
- Scheduling a workflow that only makes sense once (a single-account research job for a deal that closes next week).
- Running enrichment on the same list nightly — waste. Enrich on new-only, or weekly-delta.
- Cadence shorter than the data-refresh clock (nothing about posts or titles moves hourly).
- No cost ceiling. Every schedule must have one or it's a time bomb.
- Email delivery for workflows that produce 1000+ rows. Write to a file or CRM instead.
- Scheduling during a product launch or on-call week without a pause mechanism.
Handoff patterns
After scheduling, always remind:
- "You can list all schedules by asking me to
/schedule list."
- "Edit by saying
update my weekly prospect refresh: change the industry to Fintech."
- "Pause by saying
pause all richapi schedules for 2 weeks."
- "Cancel by saying
delete the weekly prospect refresh schedule."
Cost
This skill does not spend Richapi credits. It only sets up future spends — which are gated by the ceiling the user agreed to.
