Claude gets hands.
Up to this point, every Skill you've built produces text. You read it, you approve it, then you do the work of carrying it where it needs to go. This unit ends that.
Phase 5 is the bridge from advice to action. Through standard integrations — Model Context Protocol (MCP) and similar tools — Claude can reach into Gmail, Drive, your calendar, Slack, your CRM. The Skills you built in Unit 4 don't get thrown away. They get extended: the Skill that used to produce text and stop now produces text and sends it. Or files it. Or schedules it. Or posts it.
This is also the unit where the stakes get real for the first time. A bad answer in Phase 1 wastes your time. A bad action in Phase 5 sends the wrong email to the wrong client. So the unit's other job is to teach the single most important habit in the autonomy zone: confirm before act.
What you'll walk out with
A clear map of which tools to connect first and which to wait on. A four-move pattern (R-R-C-A) for designing Skills that act safely. Two worked examples of professionals who extended their Phase 4 Skills with action. And two hands-on designs of your own — including a "confirm-before-act" rulebook tailored to your situation.
What Unit 5 covers
- The courier problem — why even great Skills leave you doing manual carrying.
- The mindset shift — from text producer to operator.
- R-R-C-A — Read, Reason, Confirm, Act. The four moves of an operator Skill.
- The integration map — read tier, act tier, caution tier. What to wire up first.
- Two worked examples — Marcus's pipeline recap that posts itself; Esteban's Monday Skill that sends emails directly.
- Three Phase 5 mistakes — connecting too much too fast, skipping confirm, mixing systems.
- Two hands-on builds — extending a Skill with R-R-C-A, designing your own confirm-before-act rulebook.
- The Phase 5 ceiling — single-task orchestration, and the hook into Unit 6.
The courier problem.
You triggered the Skill. Claude produced a flawless output. Then you spent the next three minutes carrying that output into Slack, into Drive, into your inbox, into your calendar. That carrying is the Phase 4 ceiling.
For working professionals deep into Phase 4, the courier problem shows up in three predictable ways. Watch yourself for a day — you'll see all three.
1. The copy-paste tax.
Every Skill output ends with a copy-paste into a real tool. The Friday pipeline recap that Marcus drafts gets copied into Slack — header by header. The subject lines Maria generates get pasted into her email platform. The renewal brief Linda writes goes into the CRM. The Skill saved the writing; it didn't save the moving.
2. The forgotten last step.
You finish the Skill, get distracted, never actually send the email. The Slack post goes up Tuesday instead of Friday. The calendar invite waits in a draft tab for three days. Every "carrying" step is a place forgetfulness can leak — and it's never the work that gets forgotten, it's the delivery.
3. The "Claude should just do this" thought.
You finish reviewing the output. You know exactly where it should go. You start the manual move. Halfway through, you think: "Why am I doing this? Claude already knows where it goes." That thought is the entire setup for Phase 5.
The honest version
You can stay in Phase 4 forever and be more productive than 95% of working professionals. You can also reclaim another 30 to 60 minutes a day by stopping the carrying — without giving up the control that makes you trust the work.
Reflection · 2 minutes
Pick one Skill from Unit 4. What's the FIRST thing you do by hand after the Skill produces its output? (Paste it somewhere? Forward it? Schedule it?) That's your first Phase 5 candidate.
From producer of text to operator in your systems.
The shift sounds small. A teammate who can produce text becomes a teammate who can also send the email. The friction of every workflow drops by half — and the role of your judgment changes.
For the first four phases, your judgment showed up at the start. You decided the task. You wrote the brief or triggered the Skill. Claude produced output. You used the output. Phase 5 moves your judgment closer to the end of the workflow — you're approving an action right before it happens, instead of composing the request right before it runs.
The new-employee analogy, sharpened
Day 1 you briefed your colleague. Week 4 you delegated tasks to them. Month 3, they got system access — email, calendar, the CRM, the shared drive. You stopped being a middleman in their workflow. You also started watching what they did with the access, especially the first few times. That watching is exactly the habit Phase 5 calls for, and it has a name: confirm before act.
The slogan that does the work
Claude proposes the action. You approve it. Then it happens.
The autonomy zone (Phase 5+) only works if this rule is built in by default. Confirm before act is non-negotiable for the first month of any new integration. After that, you can loosen it where the consequences are reversible and you've built trust.
What your interactions will look like
Before: "Run the pipeline recap." Claude writes it. You copy. You format. You paste into Slack. You tag the right people. You hit send.
After: "Run the pipeline recap and post it." Claude pulls from the CRM. Writes the recap. Shows it to you with the destination ("ready to post to #leadership-revops"). You approve. Claude posts.
Same amount of your judgment. None of the carrying.
R-R-C-A — the four moves of an operator Skill.
Every Phase 5 Skill extends the four parts of a Phase 4 Skill (T-I-P-O) with four operating moves: Read, Reason, Confirm, Act. Get all four right and the Skill is safe, fast, and reliable.
Read
Claude reaches into your tools to get the inputs it needs. The CRM export. The customer's last email. Yesterday's calendar. The contract in Drive. Inputs that used to live in your head (or in your copy-paste buffer) come from the source directly.
Reads are the safest integration to start with — Claude is observing, not acting.
Reason
Claude composes the output — the recap, the draft email, the calendar invite — using the inputs and the procedure from your Phase 4 Skill. This is the part that hasn't changed: the procedure runs, the output gets shaped according to your spec.
Confirm
Before any irreversible or visible-to-others action, Claude shows you what it's about to do. The draft email and the recipient. The Slack post and the channel. The calendar invite and the attendees. You approve, you redirect, or you cancel.
Confirm is the single most important habit in the autonomy zone. Skip it and you're not in Phase 5 — you're in danger.
Act
On your approval, Claude executes — sends the email, posts the Slack message, files the doc, schedules the meeting. You stop being the courier. The full chain from data to delivery runs inside one Skill.
The integration tiers
Not every tool is equal-risk to wire up. A useful mental model: three tiers, connected in order.
Tier 1 — Read-safe
- Calendar read
- Files / Drive read
- Inbox read (search, summary)
- Slack channel read
Tier 2 — Act-reversible
- Draft email (not send)
- Save to Drive
- Slack message to your own DMs
- Calendar tentative hold
Tier 3 — Act-irreversible
- Send email to external contacts
- Post to a team Slack channel
- Schedule and invite a meeting
- Write into the CRM
Not yet
- Anything that costs money
- Anything that can't be undone
- Anything legally binding
- Personal accounts (banking, medical)
The connect-order rule
Read-safe first. Act-reversible second. Act-irreversible only when you've watched the first two for a month. The pace separates a thriving Phase 5 setup from a regrettable one.
Marcus's pipeline recap, now with hands.
In Unit 4 Marcus built a Skill that produced his weekly pipeline recap — but he still copied it into Slack manually. This unit's job is to extend that Skill so the whole loop runs end-to-end.
His Phase 4 Skill stays. He layers R-R-C-A on top:
Pull from connected tools:
- Salesforce: this week's pipeline export (opportunity stage, ACV, last activity date)
- Salesforce: closed-won and closed-lost deals from the last 7 days, with reason codes
- Calendar: any deal-related meetings on the books for next week
Run the original Phase 4 procedure: WoW change, three biggest movers, stalled-deal watch list, top-three loss reasons, leadership takeaway. Compose the 200-word Slack-ready writeup matching last week's structure.
Show Marcus:
- The draft message text
- The destination channel (
#leadership-revops) - Anyone who'll be @-mentioned
- Anything new or unusual the Skill noticed (e.g., a deal that broke the watch-list threshold)
Marcus replies send, edit, or hold.
On send:
- Post the message to
#leadership-revops - Save a copy to
/sales-ops/weekly-recaps/2026/in Drive, named with the Friday date - Report back: "Posted at 3:08pm. Filed to Drive. Ready for next week."
What changed for Marcus
Before: 45 minutes Friday, mostly writing. Then 5 minutes carrying. Then forgetting to file half the time.
After: 5 minutes Friday, all review. Confirm-or-edit. Post. Filed. Forgotten by 3:15pm. The Skill stopped at "produce text" before. Now it stops at "delivered."
Esteban's Monday morning Skill.
Esteban runs four coffee locations. Every Monday at 8am he reviews the weekend's numbers, drafts an email to his location managers, and schedules a Tuesday call with whichever location had the worst weekend. He's done this for two years.
His Phase 3 setup is a Project called "Operations" with last quarter's P&L, his location-by-location standards, and a writing voice guide. His Phase 4 Skill produced the weekly recap email but stopped there. Now he wires it for action:
Pull this weekend's numbers from connected tools:
- Square POS: per-location sales, transaction count, ticket size, top three SKUs
- Labor tracker: hours worked per location vs. plan
- Last Monday's recap email (for tone and structure continuity)
Compose the weekly recap. Identify which location had the largest negative variance vs. plan. Draft a short, warm email to the four location managers covering: weekend headlines, the one location flagged, and Esteban's read on the cause.
Show Esteban:
- The draft email text and recipient list
- Which location was flagged and why
- A proposed calendar hold for a 15-minute call with the flagged location's manager on Tuesday between 9–11am
On approval:
- Send the email from Esteban's Gmail to all four managers
- Send a calendar invite to the flagged location's manager for the 15-min call
- File a copy of the recap into Drive under
/operations/weekly-recaps/
Why this Skill is safer than it looks
Three things Esteban built in that look small and aren't:
- Email recipients are fixed. The Skill always sends to the same four managers. Claude never picks recipients.
- Calendar invite is a hold, not a confirmation. The manager gets the proposal; Esteban already knows the time window is open.
- Confirm step shows the flagged location. If Claude's reasoning is wrong (a one-off equipment failure rather than a real performance issue), Esteban catches it before the email goes out.
Skill saves Esteban about 25 minutes every Monday. More importantly, it never forgets to file the recap. The boring step that always got skipped now happens.
Reflection · 3 minutes
Take the Skill you sketched in Unit 4. Walk through R-R-C-A. What does Claude need to read? What's the Confirm step show you? What's the Act step actually do?
Three Phase 5 mistakes that turn helpful Skills into expensive ones.
Phase 5 is the first phase where mistakes have real consequences. These are the three patterns that cause them. Each one looks like progress at the time.
Mistake 1 — Connecting too much too fast.
You discover MCP, get excited, and wire up Gmail + Calendar + Slack + CRM + Drive all in the same week. Each one adds risk and complexity. By the end of the week you don't know which integration is causing the weird behavior and you start unplugging things at random.
Mistake 2 — Skipping confirm.
The Skill works. The first ten runs were perfect. You remove the Confirm step to save the 30 seconds it costs. Run eleven goes to the wrong client. There is no Phase 5 mistake more common, more expensive, or more avoidable.
Mistake 3 — Mixing personal and work systems.
You connect your personal Gmail to the same setup as your work Gmail "just to test it." A Skill that was supposed to draft a client email pulls context from your weekend texts and quietly leaks something into the draft. By the time you catch it, the email is sent.
The Phase 5 debugging move
When a Phase 5 Skill misbehaves, walk through R-R-C-A. Which move drifted? The Read pulled the wrong data. The Reason used the wrong template. The Confirm step showed the wrong destination. The Act went to the wrong place. Almost every failure traces to one of those four — and the fix lives inside the Skill spec, not in the next prompt.
Your turn.
Two hands-on builds. First you extend a Phase 4 Skill with R-R-C-A. Then you write your own confirm-before-act rulebook — the standing rules that apply to every operator Skill you ever build.
Extend your Skill with R-R-C-A.
Take a Skill you designed in Unit 4. Layer R-R-C-A on top. The Read pulls inputs from real tools. The Confirm step shows you exactly what's about to happen. The Act delivers.
One way to extend a customer renewal Skill
Read: Salesforce (account's renewal date, ACV, last 90 days product usage, open support tickets). Gmail (last three emails from anyone at the customer's domain). Drive (most recent QBR notes if a file exists matching the customer name).
Reason: Run the original renewal-risk procedure — score risk Low/Medium/High, surface drivers, recommend a move, draft an opening line.
Confirm: Show me the risk score, the top two drivers, the recommended move (call/email/exec touch/no action), the draft opener, AND the proposed outreach recipient (the contact name from Salesforce). Highlight if the contact has changed since last QBR.
Act: On send: send the draft from my Gmail to the contact, save a copy of the renewal brief to /customer-renewals/2026/ with the customer name, and add a follow-up task to my Salesforce queue dated 7 days from now. Report back the timestamp.
Notice the safety belts. The recipient is pulled from Salesforce (not invented). The Confirm shows the recipient AND highlights changes. The Act creates a follow-up task — so if the customer doesn't reply, the Skill won't quietly disappear into the void.
Write your confirm-before-act rulebook.
A one-page rulebook covering every Skill you'll ever build. What requires confirmation, what doesn't, what's never automated. Write it once; reference it every time you build an operator Skill.
Your operator rulebook.
Working professionals who skip this step build Skills that work the first month and bite them in month two. Spend ten minutes writing your rules.
One way to write your rulebook
Always confirm: Sending external email of any kind. Posting to any team Slack channel I don't own. Sending calendar invites to people outside my company. Writing into any CRM record. Saving to any Drive folder shared outside my team.
Skip confirm: Filing to my own Drive folders. Drafting (not sending) email. Saving a tentative calendar hold to my own calendar. Posting to my own Slack DMs. Reading anything.
Never automate: Anything that costs money. Anything legally binding. Anything related to compensation, hiring, or termination. Sending email to my board, my CEO, or any regulator. Any action on a personal financial or medical account.
Personal/Work line: Work Claude account is connected to: work Gmail, work Calendar, work Drive (corp domain only), work Slack, Salesforce. Personal Claude account is connected to: personal Gmail and personal Calendar only. Never the same Claude session, never the same Project, never the same Skills.
Print this and keep it next to your monitor. When you build a Skill, walk through it before you ship. Save yourself the month-two regret.
Still one task at a time.
Phase 5 changes the relationship between you and Claude's outputs. It doesn't yet change the relationship between you and Claude's workflow. You're still the conductor.
Watch yourself for a week running Phase 5 Skills. You'll notice a pattern: each Skill is end-to-end, each Skill is one task. But your real work isn't one task at a time. Your real Monday morning is three or four interleaved tasks — pull the numbers, draft the email, schedule the call, file the recap, post the summary, escalate the issue. Right now you're triggering each one separately.
What Phase 6 fixes
Phase 6 — Trust — lets you hand off the whole Monday morning, not one task at a time. You give Claude the objective (not the click-by-click) and a clear escalation rule. Claude designs the sequence, runs the steps, pauses where you said to pause, and reports back when done.
The work you just did in Phase 5 — your operator Skills, your confirm-before-act rulebook — becomes the building blocks. Phase 6 chains them.
The signal you're ready for Phase 6
You're triggering multiple Skills back-to-back to do the same thing every Monday. You know the sequence. Claude knows each Skill. You just want to say "run the Monday morning routine" once and review at the end.
Reflection · 2 minutes
What's the sequence of Skills (or recurring tasks) you'd most love to hand off as one chain? Name three steps in order.
Are you ready to move to Trust?
Three questions to diagnose your readiness for Phase 6 — chained workflows that hand off entire routines, not single tasks.
1. How many of your Skills currently touch your real tools?
2. How do you handle the "carrying" step today — moving Claude's output into the right tool?
3. How often do you find yourself triggering 2–3 Skills back-to-back to complete one routine?
Your readiness
Answer all three questions above to see your readiness for Phase 6.
Seven questions.
One per major idea in this unit. Click an answer for instant feedback.
0 of 7 answered · 0 correct so far
1. The Phase 4 ceiling — the reason your operator-quality Skills still leave you working — is:
2. The four moves of an operator Skill are:
3. The single most important habit for any Phase 5 Skill is:
4. Which tier of integration should you connect FIRST when starting Phase 5?
5. Mixing personal and work Claude integrations is dangerous because:
6. The most common Phase 5 mistake — and the one that quietly breaks trust — is:
7. The Phase 5 ceiling — the limit even great operator Skills can't push through — is:
Finish the knowledge check to see your score.
Your R-R-C-A sketch and your operator rulebook are saved — they're real artifacts you can use this week.
The new thought process.
Claude gets hands.
Stop being the courier between Claude and your real tools. Operator Skills do the full loop — read, reason, confirm, act.
Confirm before act.
The single most important habit in the autonomy zone. Never skip it for irreversible or external actions.
Connect in order.
Read-safe first. Act-reversible second. Act-irreversible only when you've trusted the first two for a month.
The sequence is next.
Your Phase 5 Skills are single tasks. Unit 6 chains them — Claude takes a whole routine and runs it end-to-end.