Contenu du pack

Un fichier SKILL.md (le guide pratique) + un modèle de départ. Aperçu :

Proactive Morning Brief Designer

The idea in one paragraph

A proactive agent is only as good as its brief. If it surfaces everything, you stop reading it; if it surfaces too little, you stop trusting it. This skill designs the brief before the agent writes one: which sources it may read, how it decides what counts as urgent, what it is allowed to draft but never send, and how a single correction ("not this, ever") retrains its ranking.

What it is built on

Proactive agents (Anthropic's Claude Duet, OpenAI's Dots, Meta's Muse) check connected apps on a schedule — Duet sends a morning brief, then re-checks every few hours — and ask approval before sending, paying, signing, or deleting. That schedule is the design constraint: anything urgent can sit until the next check-in, and anything the agent misunderstands becomes a draft you did not ask for. The brief must therefore say what it checked, when it last checked, and how sure it is — not just what it found.

Workflow

1. Source charter (input: your apps -> output: a read list with reasons).
   List every connected app and mark it Brief / Draft-only / Never read.
   Email and calendar are usually Brief; payments are Draft-only (invoice
   reminders drafted, never sent without approval); personal messages are
   Never read unless you explicitly opt them in. Write one sentence per
   source explaining why — the charter is what you show a colleague who asks
   "what does it see?"
2. Urgency ladder (input: last week's real items -> output: 4 rungs).
   Sort 20 real items from last week into: Act today / Act this week /
   FYI / Never surface. The ladder must have a test for each rung, e.g.
   "Act today = money moves, a person is waiting on me, or a deadline is
   inside 24 hours." Anything that fails all tests is FYI at most. If a rung
   holds more than 7 items on a normal day, your tests are too loose.
3. Brief template (input: ladder -> output: the fixed shape).
   Every brief has exactly five blocks, in order: (a) Needs you today (max
   5 items, each with source + why it is urgent), (b) Drafts waiting for
   approval (never auto-sent), (c) Done while you were away, (d) Missed
   check-ins (apps it could not reach, with times), (e) One-line confidence
   note ("checked 06:00–06:40; Slack unreachable"). No block may grow
   without shrinking another.
4. Correction loop (input: your edits -> output: a rule, not a grudge).
   Every time you demote, dismiss, or rewrite a brief item, turn it into a
   written rule: "invoices under $50 are FYI, not Act today" or "never
   surface social mentions before 9am." Keep the rule list under 20; prune
   monthly. A brief that cannot explain its own ranking is a brief you
   will stop reading.
5. Approval gates (input: action types -> output: send/pay/sign/delete matrix).
   For each action type, write: auto-allowed / draft-for-approval / never.
   Default: sending to anyone outside your team, any payment, any signature,
   and any deletion are draft-for-approval or never. Test the matrix with
   three uncomfortable examples before you switch the agent on.

Aperçu du modèle

Morning Brief Worksheet

Fill this in with your AI, one section at a time. The output is your brief spec.

1. Source charter

2. Urgency ladder

• Act today (test):
• Act this week (test):
• FYI (test):
• Never surface (test):

3. Brief shape (5 blocks, max items per block)

1. Needs you today (max 5):
2. Drafts waiting:
3. Done while away:
4. Missed check-ins:
5. Confidence note:

4. Correction rules (keep under 20)

1. 2. 3.

5. Approval matrix

Le pack de compétence lui-même (SKILL.md + modèle) est rédigé en anglais.

Acheter le pack complet — $4.99 Pro — les 511 compétences