routine-setup

When to invoke this skill

Takes something the person does over and over — a Monday report, a morning check, a monthly export — and makes it run on its own. Use when asked to "automate this", "do this every week", "schedule this". Grit's skill; the weekly look back over every routine is /routine-report.

The hard boundary

• Seen once before it is scheduled. The steps run once, now, with the person watching, and they say the result is right. Only then is the schedule created.
• A schedule can only do what a yes already covers. A routine that would send, post, publish, pay, delete or sign in on its own is not scheduled as such: those steps become "draft it and tell the person", and the person approves each item when it comes up. Say this plainly when the task includes one.
• No secrets in the routine. No passwords, tokens or keys written into the steps, scripts or log. A step that needs a login uses a session the person already has, or the routine stops and says so.
• The runtime's own scheduler. Use the agent runtime's schedule tool (localoy schedules, or the runtime's equivalent). No scheduler → write the routine and say how to run it by hand; never install system cron jobs or launch agents without the person's yes to that exact entry.

What you need first

• The task, and how often (and when — day, time, timezone).
• What a good result looks like — the file, the number, the message.
• Who should hear about it, and when — default: the person, only when something changed or failed.

Ask for what is missing in a SINGLE message, then wait.

Procedure

1. Write the steps. Watch or ask how the person does it now; write each step as something a fresh agent can follow with no memory of this chat: inputs (paths, URLs), the skill to call when one exists (/competitor-watch, /money-report, /qa-only…), what to write, and what counts as "changed" and "failed". Slug: <routine> (monday-competitors).

Scratch for this run lives in .localstack/work/{date}-{slug}/ — hidden, one directory per run, so a new run never clobbers an earlier one and the folder's top level stays the standard files. Scratch is disposable; old run directories may be deleted freely.

2. Write PLAN-routine-<routine>.md:

# Routine — <routine>
Schedule: <every Monday 09:00 Asia/Dhaka> · Owner: <person>
Good result: <one line> · Tell: <who, when>

## Steps
- [ ] steps written
- [ ] run once with the person
- [ ] scheduled

## Routine steps
1. …

## Changed means / Failed means
- changed: <…>
- failed: <…>

## Run log
| When | Result (ok/changed/failed) | What happened | Output |

3. Run it once, together. Follow the steps exactly as written — not from memory — and show the result. Fix the steps (not just the run) for anything that went wrong, and run again until the person says it is right. Log each trial run in ## Run log.

4. Schedule it. Create the schedule with the runtime's schedule tool, pointing it at this plan ("follow PLAN-routine-<routine>.md, append one row to its Run log"). Record the schedule id and the next run time in the plan. Show the person the exact schedule before creating it.

Standard files. This folder is kept in files any agent already reads. Update them in place; never scatter output into new folders.

• AGENTS.md — create it if missing. localstack owns only the block between <!-- localstack:start --> and <!-- localstack:end -->; rewrite that block, never anything outside it. The block says what this folder is for, the rules (drafts only; nothing is sent without the user's explicit yes, one message at a time; no invented facts), a map of the files below, and one line per topic (its PLAN, its lead count, the next unticked step) and per report (its file and date).
• CHANGELOG.md — create it if missing (# Changelog). Add one bullet for this run under today's ## YYYY-MM-DD heading, newest date first: the skill, the topic, and the counts or outcome (e.g. - lead-search austin-dentists: 18 found, 3 skipped as already contacted).
• TODOS.md — create it if missing (# TODOs). Add each open next action as - [ ] <action> (<topic>); tick items this run finished; never delete lines.
• DESIGN.md — decisions meant to last (positioning, tone, channels to use or avoid). Read it before writing anything a person will see; add to it only when the user states or approves a decision.

For this step: tick the plan's own Steps as each is done. The CHANGELOG line gives the routine, its schedule and its first run's result. TODO: - [ ] review the first scheduled run of <routine> (routine-report).

5. Report. The schedule, the next run time, where its log is, and what it will tell the person and when. Offer /routine-report for a weekly look across routines.

Completion status. End the chat report with one of:

• DONE — completed, with the evidence named (files written, counts, URLs).
• DONE_WITH_CONCERNS — completed, and list each concern.
• BLOCKED — cannot proceed; say what blocked it and what was tried.
• NEEDS_CONTEXT — missing information; say exactly what is needed.

Quality bar

• Nothing scheduled that has not run once in front of the person.
• The steps work for an agent that has never seen this chat.
• Every run leaves a row in the Run log, ok runs included.