Schedule Recipe Skill
Convert a recurring intent — "competitive briefing every Monday 8am" — into the concrete, copy-paste setup for whatever runner the user actually has, with failure handling so it degrades loudly, not silently.
What This Skill Produces
- A runner recommendation (or confirmation of the user's choice) with the reason
- The exact setup — command, cron expression, or workflow file — ready to paste
- The run prompt: what the scheduled agent should do each cycle, including which skill to load
- Failure alerting and a first-run test plan
Required Inputs
Ask for (if not already provided):
- What should run — which skill or task, and what inputs it reads each cycle
- Cadence and timezone — "every Friday 4pm" means nothing without one
- Where it can run — Claude Code (routines/loops), a server with cron, n8n, or GitHub Actions
- Where the output should land — file in a repo, Slack/email, a Brain folder, a PR
Runner Selection
Pick the simplest runner the user already has, in this order:
| Runner |
Choose when |
Setup shape |
Claude Code routine (/schedule) |
The user lives in Claude Code and the task needs an agent (reads repos, runs skills) |
A scheduled cloud agent with the run prompt |
Claude Code /loop |
Same-session polling or short-lived recurrence, not a standing schedule |
/loop <interval> <command> |
| GitHub Actions cron |
Inputs and output both live in a repo; team wants runs versioned and reviewable |
A workflow YAML with schedule: trigger |
| n8n / Make |
The trigger or output is a SaaS app (Slack, CRM, sheets) |
A workflow calling the skills REST API |
| System cron |
A server exists and the task is a script |
A crontab line invoking the CLI |
State the choice and the runner-up. If the user names a runner, use it — don't relitigate.
The Run Prompt
Every recipe includes the prompt the scheduled run executes. It must contain:
- The skill to load and the artifact to produce
- The sources to read this cycle — explicit paths/URLs, not "the usual"
- Where to write the result and how to mark the edition (date, sources read)
- What to do on missing sources — name the failure behaviour, never fabricate
- Delta instruction if recurring: read the previous edition first and report changes (see
delta-briefing)
Output Format
Schedule Recipe: [task] — [cadence]
Runner: [choice] — because [one line]. Runner-up: [alternative] if [condition].
Setup (copy-paste):
[the exact command / crontab line / workflow YAML / n8n outline]
Run prompt:
[the full prompt the scheduled agent executes each cycle]
Failure alerting: [how a failed/skipped run becomes visible — e.g. the run posts an error note to the same channel it would post the brief].
First-run test: trigger one run manually now; check [the two or three things that prove it worked] before trusting the schedule.
Quality Checks
Anti-Patterns
1---2name: schedule-recipe-23description: "Turn 'run this every Friday at 4pm' into a working, copy-paste schedule on the user's actual runner. Use when asked to schedule a recurring AI task, set up a routine or cron job for a skill, automate a weekly report, or wire a skill into n8n or GitHub Actions. Produces the exact setup for the chosen runner plus the prompt to run, failure alerting, and a first-run test plan."4homepage: https://mohitagw15856.github.io/pm-claude-skills/skill/schedule-recipe.html5metadata:6 source: https://skillmd.com/skills/mohitagw15856/schedule-recipe-27 openclaw:8 emoji: 🛫9---1011# Schedule Recipe Skill1213Convert a recurring intent — "competitive briefing every Monday 8am" — into the concrete, copy-paste setup for whatever runner the user actually has, with failure handling so it degrades loudly, not silently.1415## What This Skill Produces1617- A **runner recommendation** (or confirmation of the user's choice) with the reason18- The **exact setup** — command, cron expression, or workflow file — ready to paste19- The **run prompt**: what the scheduled agent should do each cycle, including which skill to load20- **Failure alerting and a first-run test plan**2122## Required Inputs2324Ask for (if not already provided):25- **What should run** — which skill or task, and what inputs it reads each cycle26- **Cadence and timezone** — "every Friday 4pm" means nothing without one27- **Where it can run** — Claude Code (routines/loops), a server with cron, n8n, or GitHub Actions28- **Where the output should land** — file in a repo, Slack/email, a Brain folder, a PR2930## Runner Selection3132Pick the simplest runner the user already has, in this order:3334| Runner | Choose when | Setup shape |35|---|---|---|36| **Claude Code routine** (`/schedule`) | The user lives in Claude Code and the task needs an agent (reads repos, runs skills) | A scheduled cloud agent with the run prompt |37| **Claude Code `/loop`** | Same-session polling or short-lived recurrence, not a standing schedule | `/loop <interval> <command>` |38| **GitHub Actions cron** | Inputs and output both live in a repo; team wants runs versioned and reviewable | A workflow YAML with `schedule:` trigger |39| **n8n / Make** | The trigger or output is a SaaS app (Slack, CRM, sheets) | A workflow calling the skills REST API |40| **System cron** | A server exists and the task is a script | A crontab line invoking the CLI |4142State the choice and the runner-up. If the user names a runner, use it — don't relitigate.4344## The Run Prompt4546Every recipe includes the prompt the scheduled run executes. It must contain:471. **The skill to load** and the artifact to produce482. **The sources to read this cycle** — explicit paths/URLs, not "the usual"493. **Where to write the result** and how to mark the edition (date, sources read)504. **What to do on missing sources** — name the failure behaviour, never fabricate515. **Delta instruction** if recurring: read the previous edition first and report changes (see `delta-briefing`)5253## Output Format5455### Schedule Recipe: [task] — [cadence]5657**Runner:** [choice] — because [one line]. Runner-up: [alternative] if [condition].5859**Setup (copy-paste):**60```61[the exact command / crontab line / workflow YAML / n8n outline]62```6364**Run prompt:**65> [the full prompt the scheduled agent executes each cycle]6667**Failure alerting:** [how a failed/skipped run becomes visible — e.g. the run posts an error note to the same channel it would post the brief].6869**First-run test:** trigger one run manually now; check [the two or three things that prove it worked] before trusting the schedule.7071## Quality Checks7273- [ ] The cron expression / schedule matches the stated cadence *and timezone* — show the conversion74- [ ] The setup block is genuinely copy-paste: no `<placeholders>` left except secrets, which are named75- [ ] The run prompt names explicit sources and an output destination76- [ ] A failed run is visible somewhere a human already looks77- [ ] The recipe includes a manual first-run test, not just "it'll fire Monday"7879## Anti-Patterns8081- [ ] Do not pick a runner the user doesn't have — a perfect n8n flow is useless without n8n82- [ ] Do not write a run prompt that says "as usual" or relies on the agent remembering prior runs without a stored previous edition83- [ ] Do not schedule without failure alerting — silence and success must not look identical84- [ ] Do not default to hourly/daily to "be safe" — match the cadence to how often the inputs change85- [ ] Do not put secrets inline in the setup block — reference the runner's secret store