Command Center — starter

Your own dashboard, running only the sections you install. Nothing here reads anyone else's data: your notes live in this folder, and the tokens are yours.

Four sections are wired up out of the box:

Tab	What it does	Reads / writes
🏠 Home	Empty page, yours to fill	—
🔀 Pull requests	Per-squad PR checkpoints from GitHub, Slack channel signals, and a ready-to-paste Slack draft	.data/
✅ Tasks	Eisenhower-quadrant backlog + a Today block	content/tasks/active.md
📓 Journal	Per-day journal	content/tasks/journal.md
📅 Meetings	Meeting notes by bucket, with live action-item checkboxes and archiving	content/meetings/

Setup (~2 minutes)

1. Install. The section packages are public on npm — no token, no registry config, nothing to authenticate:

pnpm install

Upgrading from an early copy of this template? Delete any .npmrc in this folder first. An earlier version told you to point the @asucregonzalez scope at GitHub Packages; the packages now live on npm, and a leftover .npmrc sends the install to the wrong registry and 404s. .npmrc is gitignored, so git pull can't remove it for you:

rm -f .npmrc && pnpm install

2. Configure.

cp .env.example .env

• GITHUB_TOKEN — required by Pull requests. Classic token with repo scope, so it can read PRs in the orgs you track.
• SLACK_BOT_TOKEN — optional. Only for channel signals and posting the draft.
• VITE_OWNER_NAME — your name as it appears in an Owner: field, so Meetings can split action items into yours vs everyone else's.

Tasks, Journal and Meetings need no tokens at all — they're just your markdown.

3. Run.

pnpm dev          # or: make dev

make on its own lists every shortcut — make setup does the one-time bootstrap (isolated Google Workspace config dir + local settings), make run launches Claude with this workspace's environment.

Open http://localhost:5273. The API runs on :4320 and the dev server proxies /api to it.

Pull requests starts with no projects — add your first squad from its own UI (a name plus the GitHub logins to track). Tasks, Journal and Meetings start from the seeded files in content/, which double as format documentation.

The skills

.claude/skills/ ships eight Claude Code skills. These five keep the dashboard's data current, so you're not hand-editing markdown:

Skill	What it does
task-management	Creates and edits tasks in content/tasks/active.md, enforcing one action per task
daily-briefing	Reconciles yesterday, gathers calendar + mail + Slack + PR queue + backlog, proposes today's Top 3, writes the Today block and archives yesterday to the journal
meeting-processor	Pulls yesterday's meetings from Granola into content/meetings/, and their action items into your task list
sync-meetings	Same, but from a Google Drive folder — use whichever matches where your notes live
google-workspace-cli	Reference for the gws CLI, which the two Google-backed skills above rely on

Each needs a few blanks filled in before first use — your GitHub login, Slack member ID, Granola folder names or a Drive folder id. Every skill says so at the top. A skill whose tool you haven't authenticated skips that source and tells you, rather than making something up.

The two meeting skills are the ones you'll hit first: both start with no source configured, so they'll ask instead of syncing. That's deliberate — a bucket is a decision about how you organise your work, not something to guess. Answer with a Drive folder id (sync-meetings), your Granola folder names (meeting-processor), or "neither, I'll write notes by hand" — the Meetings tab reads content/meetings/*.md either way.

Ask for them in plain language ("what's my plan today?", "process yesterday's meetings") or invoke directly with /daily-briefing.

A second brain, if you want one

Four more skills turn content/memory/ into a wiki that compounds:

Skill	What it does
wiki	Drop a document in content/sources/, ask to ingest it, and it becomes linked pages in content/memory/. Also queries and lints the knowledge base
memory-claude-md-sync	Fires on every wiki write so the navigation hub never drifts from reality
claude-md-template	Keeps every CLAUDE.md to one shape, so the hierarchy stays navigable
/weekly-done-cleanup	Prunes completed tasks out of the backlog into content/tasks/done-archive.md, keeping ones that still give context

And these are general-purpose, nothing to do with the dashboard:

Command / skill	What it does
/setup	Walks you through first-time setup — identity, placeholders, isolated Google auth, git history
/wiki-ingest	Ingest one document into the wiki, proposing what to write before writing it
/grill-me	Interrogates a plan branch by branch until the decisions are actually resolved
/compact-session	Captures a long chat's context into notes plus a resume prompt for a fresh session
create-presentation	Builds a Fever-brand deck in Google Slides — brand kit and the Slides-API gotchas already solved

content/memory/ ships empty with just its hub file. Nothing is pre-populated — it becomes useful only as you feed it.

Bootstrap scripts

/setup is an interview, not a form: it asks your name, email and GitHub handle, then your role, your current goal for the quarter, your team and timezone, and your Slack ID plus the channels worth scanning. Role and goal are the two that change behaviour rather than just filling in a blank — role picks the work-type buckets on the task board, and the goal becomes the tiebreaker the daily briefing uses when two tasks are equally urgent. Everything is optional and everything is editable afterwards (the "Me" table in CLAUDE.md).

.claude/scripts/setup/ holds the machinery it drives, all parameterized:

Script	Does
apply-placeholders.sh	Substitutes your profile — {{NAME}} / {{EMAIL}} / {{GH_HANDLE}} positionally, then any TOKEN=value pairs (ROLE, GOAL, TEAM, TIMEZONE, SLACK_ID, SLACK_CHANNELS) — across the template
apply-work-types.mjs	Bakes your role's work-type buckets into src/work-types.ts. --list shows the presets
ensure-gws-config-dir.sh, verify-gws.sh	Creates and checks an isolated Google Workspace config dir, so this assistant's auth never collides with anything else
generate-makefile.sh	Writes a makefile whose run target launches Claude with the right env
init-settings-local.sh	Seeds .claude/settings.local.json
fresh-git-init.sh	Wipes template history for a clean first commit — destructive, run deliberately

CLAUDE.md documents the workspace conventions all of these follow — worth reading once.

Recommended: rtk

rtk is a CLI proxy that compresses command output before the agent reads it — a single Rust binary, Apache-2.0. It's the cheapest efficiency win available here, because every oversized tool result gets re-read on every subsequent turn.

brew install rtk       # or: curl -fsSL https://raw.githubusercontent.com/rtk-ai/rtk/refs/heads/master/install.sh | sh
rtk init -g            # registers a global PreToolUse hook + ~/.claude/RTK.md
                       # then restart Claude Code
rtk init --show        # verify

Measured on this project's own repo:

Command	Raw	Via rtk
ls -la	1615 B	256 B (−84%)
git log -n 20	24.2 kB	6.6 kB (−73%)
git status	636 B	282 B (−56%)
git diff --stat	414 B	413 B (−0%)

Read the claims precisely, because the honest version is less dramatic than "90%":

• It compresses bash output, which is one input among your prompt, the system prompt and the conversation history — and input is only part of the bill. The reduction dilutes at each step.
• Its token figures are bytes / 4 estimates; the percentages are trustworthy, the absolute token counts are not.
• Already-terse commands gain nothing, as the git diff --stat row shows.
• Read, Grep and Glob bypass it — the hook only intercepts the Bash tool. For big files, reach for rtk read / rtk grep, or a ranged Read.

Install it globally (-g), not as a hook in this repo: a committed hook would fail for anyone who hasn't installed the binary.

Your content

content/ is a plain folder of markdown you can edit directly, in the app, or both — the sections read and write the same files.

content/
├─ tasks/active.md         ← Tasks tab. Standing quadrants + a Today block.
├─ tasks/journal.md        ← Journal tab. One `## YYYY-MM-DD` entry per day.
├─ team/task-labels.json   ← the [@label] chips Tasks can put on a task
├─ memory/CLAUDE.md        ← the wiki's hub (starts empty)
├─ sources/                ← drop documents here for /wiki-ingest
└─ meetings/
   ├─ example.md           ← 2 example meetings, with action items owned by you and others
   ├─ example-planning.md  ← a second bucket, so you can see the grouping
   └─ archive/             ← notes archived from the UI (one example inside)

Want to see the Pull requests tab populated too? It needs no token for that:

cp -R .data.example .data && make dev

That gives you one example squad and a saved checkpoint — PRs grouped by state, a Slack signal, and a ready-to-paste draft — rendered from stored JSON, so there are no GitHub calls at all. Delete .data/ when you want to add your own squad; see .data.example/README.md.

Two formats are strict, and the files show both:

• Journal entries need ## YYYY-MM-DD headings. Any other heading is ignored.
• Tasks needs **Top 3:** written exactly like that, with a numbered list under it, or the Top 3 silently won't render.

A task title can carry two leading markers, both optional and both editable from the Tasks tab:

• Priority — [P0] now · [P1] this cycle · [P2] later · [P3] nice-to-have. The board sorts by it, and an unmarked task sorts above [P3], so dropping the marker isn't a way to park something.
• Work type — [Ship], [Quality], … whatever src/work-types.ts declares. /setup generates that file from your role, because the useful buckets differ: an IC's week splits into shipping / quality / support / growth, a manager's into delivery / capacity / engineering-excellence. .claude/scripts/setup/apply-work-types.mjs --list shows the presets; the file is plain TypeScript in your repo, so edit it freely (it survives pnpm update).

Already keep notes somewhere else? Point CONTENT_ROOT at that folder in .env and the sections read it instead — it just needs the same layout.

Adding another section

pnpm add @asucregonzalez/section-<name>

Then two edits:

1. src/sections.ts — import the descriptor and add it to the array (this drives the tab bar).
2. server/index.ts — if the section has a backend, add its register…Routes to the routers list.

Layout

Path	What it is
makefile	Shortcuts — make lists them
CLAUDE.md	Workspace map + the conventions the skills and dashboard share
.claude/skills/	The skills that maintain your content and wiki
.claude/scripts/setup/	One-time bootstrap scripts
content/memory/	Your wiki — starts empty
src/sections.ts	The one file that decides which sections you run
src/work-types.ts	Your role's work-type buckets, generated by /setup — yours to edit
src/HomeView.tsx	The Home page
src/App.tsx	The shell: tab bar + the active section's view + host config
server/index.ts	Express server; mounts each section's routes
tailwind.config.js	Loads the shared theme preset, and scans installed sections for classes
content/	Your markdown (see above)
.data/	Pull-requests state: projects, squads, saved checkpoints — gitignored
.data.example/	Committed example of the above — cp -R .data.example .data to see it

Gotchas

• src/index.css: the @import '@asucregonzalez/theme/base.css' must stay above the @tailwind directives. postcss-import only inlines @import at the top of a file, so moving it down silently drops every component class and sections render half-styled.
• tailwind.config.js: keep the ./node_modules/@asucregonzalez/*/dist/**/*.js content glob. Sections ship compiled markup, and Tailwind purges any class it can't see.
• No "Regenerate" buttons? Expected. Those re-run a Claude skill on the host; this app sets refreshEnabled: false in src/App.tsx, so sections hide them rather than offer a button that can't work. If you build your own /api/refresh/<skill> route, flip it to true.
• Sections render unstyled? You're missing presets: [ccPreset] or one of the two points above.
• pnpm install edited pnpm-workspace.yaml? Expected, and harmless. pnpm 11 quarantines very recently published packages as a supply-chain precaution; when a section release is newer than that window it adds the versions to minimumReleaseAgeExclude and proceeds. Commit it or discard it, either is fine.