tm-onboard

name

tm-onboard

description

Use right after a tm-mode install (`tm-mode init` / `tm-mode join`) — when entering Claude Code/Codex in a freshly set-up team repo. Dispatches a verification subagent to confirm the install landed, and meanwhile conveys what tm-mode does for you. Triggers on "tm-onboard", "teammode onboarding", "start teammode", "is the install ok", "check teammode setup", "팀모드 온보딩", "팀모드 시작", "설치 잘 됐나", "팀모드 셋업 확인", or when the CLI tells the user to open an agent and run tm-onboard.

tm-onboard — Install Verification + tm-mode Value Briefing

The install was already completed by the CLI (tm-mode init / tm-mode join). When a person enters an agent right after setup, this skill does exactly two things:

1. Check whether the install landed correctly — delegate this to a verification subagent (the main agent does not wait).
2. Explain what tm-mode does for the person (value) — the main agent delivers this while verification is running.

⛔ Do not install or ask setup questions. Asking for member name, org, team name, role, agent, or obsidian; calling install.py directly; creating or cloning a repo — all of that belongs to the CLI wizard and has already happened. Do not reproduce it. (If the user says "set this up" while tm-mode is not installed yet → if the repo contains infra/install.py and AGENTS.md, route to the bootstrap procedure in the "첫 접촉" (First contact) section of AGENTS.md (clone-and-go — dry-run → conversational approval → that procedure handles installation). Otherwise, tell them to run "tm-mode init (new team) / tm-mode join <url> (join)" in their terminal and stop. This skill itself never runs installation in any case.)

Entry Flow (Parallel — In This Order)

1. Immediately dispatch a verification subagent (§①). Read-only; report results only. The main agent does not wait for it to finish.
2. While the subagent runs, the main agent delivers the §② value briefing to the person — do not leave them staring at an empty screen.
3. When the verification result arrives, summarize it in the user's language:
   • Everything ✅ → close with one line like "The install was also verified successfully."
   • Anything missing (❌) → explain what failed in plain language → tell the person to run tm-mode join <team-repo URL> again from the same location (install is idempotent — it safely fills in missing pieces). Do not manually reproduce the install.py steps.

Entry context: after tm-mode init/join, the CLI tells the user that opening Claude/Codex and entering 'tm-onboard' will automatically run verification and briefing (cli.py _done()). This skill is that first entry point. Assume it is running from the team root (the cloned repo).

---

① Install Verification — Delegate to a Verification Subagent

As soon as this skill starts, dispatch one dedicated verification subagent with the prompt below (read-only, absolutely no modification or installation — verification only). In dogfooding, the core bug was "hooks and skill symlinks were not registered", so do not assume success; have the subagent verify with real files and commands.

[Verification Subagent Prompt Template] — fill in <team-root>, <member-name>, and <agent>, then dispatch:

From team root <absolute team-root path>, only verify whether tm-mode was installed correctly. Absolutely no modification, installation, or git writes (read-only). Check each item below with real commands/files and report a table with ✅/❌ plus a one-line reason for anything that failed:

1. Core engine: Does python3 infra/teammode.py context --root <team-root> --json print state without errors? (state=off immediately after install is normal — installed does not mean active.) Also include the team name, member count, and session count from the output (used in the value briefing).
2. Scaffold: Is <member-name> listed in memory/team/members.md, and does memory/INDEX.md exist?
3. Team config: Does team.config.json exist, and are agents recorded?
4. Skill symlinks: In the agent skill directory (claude=~/.claude/skills, codex=the relevant path), are tm-mode skills such as tm, tm-onboard, and tm-memory symlinked/installed?
5. Hook wiring: Does the agent configuration (claude=~/.claude/settings.json) include tm-mode hooks such as session-start?

On the last line, give the overall verdict (all normal / list of missing items) in one line.

• The main agent receives only this subagent's final result and summarizes it through Entry Flow step 3. If the result is important enough to distrust self-reporting, the main agent directly re-checks only the missing items.
• One-line note for Codex users: after the first install or a hook change, they may need to open codex (TUI) once and press Trust on the hook trust prompt — otherwise hooks are silently skipped in headless mode (codex exec) (tm on detects this and reports it with [warn]).

---

② tm-mode Value Briefing (While Verification Runs)

Immediately after dispatching the verification subagent, read infra/skills/base/tm-onboard/value.md and explain the value there in human language, adapted to the person and context (founding a new team / joining an existing team, role). Do not recite it verbatim — follow the tone guide in value.md and give the key points briefly in your own words. Respond in the user's language.

💡 The single source of truth for the value content is value.md. Do not duplicate value copy in this body — if the team/founder edits only value.md, the delivered message should change.

Then add the current team status using the status returned by the verification subagent (team name, member count, session count), in the user's language.

• A brand-new team is empty (0 session logs, 0 KB items) → normal. In the value.md tone, say that the structure is in place and it starts accumulating from now — do not describe the empty state as a failure.
• Close with the next single step: say that when they start work, they should run tm on — it refreshes and opens the agent with team context. This is the only instruction needed to prevent "what now?" right after setup.
• If the team is empty (0 session logs / 0 KB items) and it looks like team documents already exist in a document service, you may add exactly one sentence: if they have existing docs, they can say "memory upload" to bring them into team memory. Mention it only; execution belongs to the tm-import-memory skill. Stop there.

---

Non-Goals / Boundaries

• No installation, setup questions, repo creation, or install.py calls — the CLI (tm-mode init/join) already handled all of it.
• Do not hold the main flow synchronously for verification — delegate to a subagent, and let the main agent deliver the value briefing in parallel (do not make the person wait).
• Do not list menus — service connection (L2), Obsidian registration, banner/personality customization, and memory upload are not handled here. When they become necessary, each skill (tm-connect, tm-customize, tm-import-memory) will naturally appear as the trigger (progressive).
• Activation (tm on) is only recommended here — actually turning it on and showing the welcome/banner belongs to the tm skill.
• Do not write code, create issues, or push.

Common Mistakes

Mistake	Correct Method
Asking again for member name, org, team name, or role	Do not ask — the CLI wizard already collected it
Making the person wait while the main agent performs verification synchronously	Dispatch a verification subagent + deliver value while it runs (parallel)
Calling install.py directly to reproduce installation	Verify only. If something is missing, guide the user to rerun tm-mode join <url> (idempotent)
Skipping verification and assuming "it must be installed"	Have the subagent verify with real files/commands — especially hooks and skill symlinks
Listing L2, Obsidian, or personality items as a menu	Do not handle them. Each skill appears when needed
Treating an empty team (0 session logs) as a failure	Normal — narrate that it starts accumulating from now
Starting installation for someone who is not installed yet	The skill does not install — inside a repo, route to the "첫 접촉" (First contact) bootstrap in AGENTS.md; outside a repo, guide them to "tm-mode init/join in the terminal" and stop

---

For behavior specs, check docs/spec/ (install.py and onboard skill); for the entry contract, check _done() in src/teammode/cli.py (it points to this skill).