Student-Designed Task — Planner (entry 5)

Module 6, Lesson 6.5 · register entry 5 · design-before-build · concept shape + recipe paths

What this is. The planner for the scheduled task you design. The only original task in the register; entries 1–4 are shared across the course. Entry 5 is yours alone.

How to use. Work through sections 0–10 in order. Finish the design before you build. A task you cannot finish on this planner is not ready to go on a schedule.

Audience = only you. No exceptions for entry 5.

If the artifact lands anywhere besides your own machine, your own drafts folder, or a private channel you are alone in, stop and redesign.

0. Choose

Pick one idea — your own, or adapted from Lesson 6.5’s list. Write it in one sentence.

Does it duplicate any of entries 1–4?

Yes / no. If yes, pick a different idea.

Does the idea respect audience = only you?

Yes / no. If no, redesign or pick a different idea; no exceptions for entry 5.

1. Why — what question does this task answer for me?

One to three sentences; be specific — “it saves me time” is not enough.

What would the first week of running this task change in my behavior?

Name at least one concrete change, one sentence.

If the task worked perfectly for six months, what would be different about my life?

2. The five moves — write each out

Plan

What I am designing (this document), what I need to decide before the first run.

Trigger

Schedule: <e.g., daily 20:00 America/New_York> / chosen scheduler: <Cowork tab (primary) / Optional advanced: cron / launchd / Windows Task Scheduler>.

Run

The agent’s job each run — one paragraph, naming the scope it reads and the artifact it produces.

Deliver

Exactly where the artifact lands. Confirm the destination is one of: file on my machine, draft in my drafts folder, self-addressed email, row in a log file, message in a private channel I am alone in.

Log

Log line format: <ISO-8601>, <task-name>, <status>, <artifact path>, $<cost>, <message>. Path: ~/ai-architect-academy/automation/<task>/log.txt.

3. Idempotency key

Key

Date / ISO week / query ID / event ID / content hash.

Why this key is stable

Filename pattern

e.g., reading-list-YYYY-WW.md.

What happens on a double-fire

e.g., “second run overwrites the same-named file; no duplicate output.”

4. Cost

Per-run estimate

Runs per month

Monthly rate (computed)

Per-run ceiling

Per-month ceiling

Effect on total-suite monthly total

New total = $<N.NN>, which is under / at / over my total-suite ceiling.

If adding this task pushes total over the suite ceiling, name the task you will retire or redesign to bring the total back under.

5. Observability

Artifact header block (exact text):

<ISO-8601 run timestamp> | <task name> | run <run-id>
Idempotency key: <key>
Searched: <sources / scopes>
Items read: <N>
Cost: $<N.NN>
Health: <one line — any scope denied, any input unavailable>
Next run expected: <ISO-8601 local>

Log line format (confirmed from section 2): as above.

Log rotation: keep last <N> days, archive older.

6. Failure modes

For each, write what the task will do:

Transient network error (timeout, 429)

Retry once with backoff, then fail loud.

Input scope unavailable (file missing, label deleted, credentials revoked)

Log error, write no artifact, leave last success to go stale.

Output shape wrong (missing section, empty where it shouldn’t be)

Write to <...>.attempt.md, do not overwrite last-good artifact.

Catastrophic (credentials permanently revoked, model deprecated)

Stop the task, write loud error, pause until human audit.

7. The prompt (draft)

Write the full prompt here. Use the shape of the other Module 6 templates as a starting point. Sections:

Output contract (specific sections, length limits)
Scope (narrow, named inputs)
Voice (one sentence of tone)
Honesty rails (anti-confabulation, zero-state behavior, limit reporting)
Delivery (exact artifact path)
Log (exact log line format)

Draft prompt

8. Register-entry fields

Copy into automation-register-v1-draft.md when you file entry 5.

Task name: <...>
Task ID: <...>
Purpose: <...>
Schedule: <...>
Scope / credentials: <...>
Idempotency key: <...>
Artifact path: <...>
Audience: me only
Per-run cost (estimate): $<...>
Monthly cost (rate): $<...>
Cost ceiling: $<...>/run, $<...>/month
Last run: leave blank until first run
Last success: leave blank
Last fire: — for non-watchers; fill in for watchers
Next review: <YYYY-MM-DD, 60–90 days from today>
Retirement trigger: <specific event>
Prompt-change log: initial entry: “YYYY-MM-DD — designed and first run.”

9. Two-run test plan

The entry-5 requirement is at least two successful runs with at least one scheduled interval between them.

Run 1 plan

On-demand trigger after build; expected artifact shape; what you will verify.

Run 2 plan

Next scheduled interval OR forced second trigger with new/different inputs; what you will verify about idempotency.

Between-run observation

One thing you will check between run 1 and run 2 — did the log line appear, did the artifact persist, etc.

10. Pre-launch checklist

The task is not a duplicate of entries 1–4.

Audience = only you — I can point to the exact destination and it is me.

The five moves are each written in full.

Idempotency key is a single specific token and the filename pattern uses it.

Monthly cost computed; ceiling set; suite-ceiling impact understood.

Observability floor is in place: header, log, next-run expectation.

Failure modes are designed for, not hoped around.

Prompt drafted with honesty rails.

Two-run test plan written, not just imagined.

When every box is checked, build and run. File register entry 5 after the second successful run.