Introduction to Claude Tag

A walkthrough of the Claude Academy course Introduction to Claude Tag (Start your journey track, 11 lessons + quiz). Each lesson is summarised with my own tables and examples; take the course for the full lessons. All the courses are listed in the Claude Academy course index.

Note: Detailed content expanded from course outline and public lesson pages; verify against lessons. Lesson-to-module grouping below is inferred from lesson order; setting names may differ by workspace/version.

Prerequisites: Claude Tag installed in your team’s (Slack) workspace; access to Claude Tag in at least one channel (some orgs restrict by role — ask your admin if tagging returns an access message). Helpful, not required: having used Claude in a chat.

Audience: Knowledge workers who recently gained Claude Tag in their channels. Admins setting it up should start with the Claude Tag admin setup guide.

Three habits that run through the course:

1. Work with it as you would a colleague — ask where you’d ask a person; give it real jobs; correct it as it works.
2. Work where the team works — keep work in the channel it belongs to so others can see, correct, continue, and so Claude’s learning helps everyone.
3. Extend trust a step at a time — build in checks; once results are reliable, let it reply untagged, run schedules, own ongoing work.

---

Module 1 — Get started with Claude Tag (Lessons 1–5)

Tag Claude to see what you can do

• Claude Tag = Claude in your team’s Slack workspace. Write @Claude + request → Claude replies in a thread, after first reading the channel (no need to re-explain background).
• With tools connected to the channel (Drive, calendar, ticketing…), it can look things up and make changes.
• Anyone in the channel can reply in the thread to add detail or redirect — Claude reads every thread reply without a new tag.
• Configure link under each Claude reply opens the channel’s settings.
• Beyond replying, it can keep working untagged:

Capability	What it means
Untagged replies	Tell Claude once which messages to handle (e.g., IT help, bug reports); it replies on its own
Standing responsibility	Looks after something over days/weeks (e.g., launch checklist)
Routines	A task described once and repeated on a schedule

• Try it: /invite @Claude in a channel, ask a question about recent work; done when the answer draws on channel content.

Claude in a DM

• Works as you: runs on your own Claude account with your connectors (calendar, email, files).
• Uses: personal work, private drafts, catching up on a named public channel (by search), personal routines (e.g., morning brief).
• Can draw on: your tools, any public channel (by search when named), attachments/pastes.
• Privacy: DM learning stays in the DM (never added to workspace notes); actions in your tools are under your name; nothing reaches team channels.
• Limits: nobody else can see/continue/correct; no access to channel-connected team tools; learning benefits only you.
• Starter questions: “What do you know about me…?” · “What can you access from here?”
• If DMs are disabled by your org, use a channel with only you and Claude.

Claude in a public channel

• Claude sees the full channel history (even before it joined) + the channel’s connections (set by whoever manages Claude Tag). By default you don’t need a personal Claude account to tag it — the org covers it.
• Examples: Sales pulls a CRM account; Data runs a warehouse query; Support drafts a ticket reply; Engineering opens a PR.
• Results land in the thread, or as a page (link) the team can revisit and Claude can keep editing.

Access type	Acts as	Who can direct it
Team connection	Claude’s own account (permissions set by admin)	Anyone in channel
Your personal connector	You	Only you

• Personal-connector prompt (visible only to you): Allow (posts directly, holds responses that look sensitive) · Allow with review (approve every response) · Don’t allow. Enterprise Owners may remove an Allow option. You can Stop a task using your connectors.
• Claude cannot read others’ private channels or DMs from a public channel.
• Try it: “what can you access from this channel?” → fetch one real record → compare to the source → correct in thread.

Claude in a private channel

• Same channel features (threads, history, connected tools, untagged replies, routines), members only.
• Tools can be connected to just this channel (least access needed); the member list is the access list — keep it tight.
• Learning saved only to this channel’s notes, never workspace notes; from elsewhere Claude behaves as if the channel doesn’t exist.
• Benefits from existing workspace notes without adding to them. Actions attributable via Claude’s own account in tool logs.
• Guest/shared (external) channels: admin setting; where enabled, Claude stays within that channel.

Pick the right place for the work

• Ask Claude where you’d ask a teammate: DM for personal, private channel for a group, public channel for team work.
• When more than one place fits, choose the public channel — others can build on it, corrections help everyone, setup serves everyone.

Place	Who sees	Tools	Where learning goes
DM	You	Your connectors	Your DM notes only
Public channel	Whole team	Channel connections (+ your connectors for your requests)	Channel notes; may create workspace notes
Private channel	Members	Member-only connections	That channel only

---

Module 2 — Shape how Claude works in your channels (Lessons 6–8)

Refine Claude’s work through memory and instructions

	Memory	Channel instructions
What	Notes Claude keeps (and you add)	Rules a person writes
Change via	Message: @Claude remember for this channel: …	Configure → settings page
Use for	Corrections, preferences, who owns what	Purpose, tone, when to reply, what never to do
Priority	Lower	Wins on conflict

• Note scopes: channel notes (that channel), workspace notes (saved only from public channels, for company-wide facts), DM notes (just you).
• Maintain memory: “what do you remember about this channel?” · “remember for this channel: [corrected]” · “that’s outdated, forget the entry about [topic].” Org Owners can also edit notes in admin settings.
• Instructions are strong guidance, not a hard lock — use settings/access controls for must-never-happen rules. Changes apply to new threads.
• Example instructions (bug-triage channel):

  This channel triages customer bug reports for the billing service.
  When a new message contains an error ID, start a thread:
  - look the error up in the monitoring tool
  - link the dashboard
  - propose an owner
  Otherwise stay quiet unless tagged.
  Keep replies under six lines; link to logs rather than pasting them.
  File confirmed bugs in the tracker with the label from-chat.
  Investigation is read-only; never touch production.

• Workflow: try a rule in memory → once proven, promote to instructions.

Symptom	Likely cause	Fix
Thin/generic replies	Missing tool access	Connect tools to channel
Answers things it shouldn’t	No clear job	Give it a job in instructions
Same correction repeats	Said once without “for this channel”	“remember for this channel”
No response	Not enabled here	Ask manager to turn it on

Proactivity: let Claude reply without being tagged

• One kind of work per channel → Claude becomes a specialist.
• Respond automatically setting lets Claude weigh each message and decide to jump in; guide it via memory:

  @Claude remember for this channel: when someone posts an account name or asks about an open item, reply even if nobody tagged you. Stay out of scheduling and social chatter.

• Good fits: support/feedback, reference/policy, project, incident channels.
• Standing responsibilities: coordination (launch trackers, reminders), owning an outcome (queue < 1 day old), continuity through change (remember decisions and why).
• Turning down noise: refine instructions first; then @Claude only respond when someone @-mentions you (channel setting) · refine scope via notes · !mute / !unmute for one thread.

Put recurring work on a schedule (routines)

• Types: daily/weekly reports, continuous monitoring, regular checks.
• One message = what to read, what to look for, what to post, when, format, and what to do if nothing found:

  @Claude every weekday at 9am, compare the onboarding tracker with what was said in this channel yesterday, and post only what needs attention… If nothing needs attention, post nothing.

• Treat the first post as a draft; reply in its thread with changes.
• Manage: “what routines do you have set up in this channel?” or @Claude !scheduled; pause/stop in plain words. Routines belong to the channel and keep running if the creator leaves.

---

Module 3 — Run one task well (Lesson 9)

Write a request Claude can work with

1. Pick the work — think outcomes not steps; confirm Claude can reach inputs; check channel visibility (sensitive → private).
2. Write the request — goal + why; process/constraints that matter; inputs/examples; a way to check (rubric, cite sources, verified vs inferred); priorities (audience, polish, decisions to bring back); result form (thread reply vs published page link).
   • Big tasks: ask for an outline/plan first. Exploratory: say so and ask for directions. Proactive work: set a pinned north star (one measurable goal).
3. Let Claude work — progress checklist edited in place (no notifications); new replies when it needs you or finishes; quiet usually = working. Editing your original message does nothing — send a new reply. Each top-level message = new task.
4. Review — follow cited links/queries; diagnose: wrong place → move/connect tool; incomplete → ask what it searched; unclear request → rewrite; missed team pattern → save to memory. @Claude !help lists thread commands.

---

Module 4 — Expand Claude’s ownership to larger tasks and ongoing work (Lessons 10–11)

Expand what Claude owns

• Building blocks recap: connections (tools) → memory/instructions (how the team thinks) → untagged replies (when to act) → routines (when to start).
• Per-channel Configure page shows these; what you can change depends on org setup.
• Your role shifts from directing tasks to setting goals and reviewing decisions. Ask Claude to group questions and bring each decision with a recommendation.
• Example ownership message:

  The goal: we launch on October 14 with no open blockers. You own keeping us unblocked until then… Get the information yourself. Bring me anything about scope, timeline, or a real trade-off, grouped twice a week.

Review your team’s habits

• Channel-by-channel checklist of habits; each gap maps to a lesson. Habits belong to the team — review together and write outcomes where Claude can read them.
• Next: Building Effective Human-Agent Teams course; best practices and use-case docs; admin customization guides.

---

Quick-Review Flashcards

1. Q: How do you start work with Claude Tag? → A: @Claude with a request in a channel; it replies in a thread after reading the channel.
2. Q: How do you add Claude to a channel? → A: /invite @Claude.
3. Q: In a DM, whose identity does Claude act under? → A: Yours, via your own connectors.
4. Q: In a team channel connection, whose identity? → A: Claude’s own account, with the access the admin gave it.
5. Q: When more than one place fits, where should you ask? → A: The public channel.
6. Q: Memory vs channel instructions — which wins? → A: Channel instructions.
7. Q: Phrase to save a channel-level correction? → A: @Claude remember for this channel: …
8. Q: Where are private-channel learnings stored? → A: Only that channel’s notes, never workspace notes.
9. Q: How do you silence one thread? → A: !mute (!unmute to reverse).
10. Q: How do you list routines? → A: Ask “what routines do you have set up in this channel?” or @Claude !scheduled.
11. Q: Does editing your original request change the task? → A: No — send a new reply in the thread.
12. Q: What’s a north star? → A: One ambitious, measurable goal a person pins in a channel so Claude can propose work toward it.

Practice Ideas

• Invite Claude to a channel and ask a question that requires recent channel context.
• Set up a personal morning-brief routine in a DM; refine it after the first delivery.
• Ask “what can you access from this channel?” and verify one fetched record against its source.
• Run one sensitive task in a private channel and confirm visibility.
• Sort four real requests into DM / private / public and justify each.
• Save a team principle with “remember for this channel”, then test in a new thread; promote it to instructions once proven.
• Write one line defining what Claude should answer untagged; test with an included and excluded message.
• Create a “post only if something needs attention” weekday routine.
• Run one task through all four steps (pick, write, let work, review) with built-in verification.
• Hand Claude an ongoing job with a north star and explicit decision boundaries; review after a few days.