Skip to content

CLAUDE.md

Latest commit

 

History

History
136 lines (112 loc) · 7.3 KB

CLAUDE.md

File metadata and controls

136 lines (112 loc) · 7.3 KB
Raw

Command Center Starter

How to Use This File

Read this file first — it maps the workspace and names the conventions the dashboard depends on. Then follow Deep-Dive Navigation to whatever you're working on.

Two things in here are strict formats, not style preferences. Break either and the dashboard silently stops showing the data:

  • content/tasks/journal.md entries need a bare ## YYYY-MM-DD heading.
  • The Today block's Top 3 needs the exact heading **Top 3:** followed by a numbered list.

Don't start with broad searches. The map below already says what exists and where.

Me

Filled in by /setup. This is the context every skill leans on to fit your situation rather than a generic one — if any of it goes stale, edit it here.
Name {{NAME}}
Email {{EMAIL}}
GitHub {{GH_HANDLE}}
Role {{ROLE}}
Current goal {{GOAL}}
Team {{TEAM}}
Timezone {{TIMEZONE}}

Role decides the work-type buckets on the task board (src/work-types.ts, generated by /setup). Current goal is the tiebreaker whenever work has to be ranked — a task that moves the goal outranks an equally urgent one that doesn't; the daily-briefing skill uses it that way when picking the Top 3.

What Is This Project

Your personal Command Center: a small dashboard shell plus the sections you chose to install, and a set of Claude skills that keep its data current.

The dashboard reads and writes plain markdown in content/, so you can edit your tasks, journal and meeting notes in the app, in your editor, or through a skill — all three touch the same files, and git is the safety net.

Tab Reads / writes
Home reads content/tasks/active.md + content/meetings/ (counts only)
Pull requests .data/ (projects, squads, saved checkpoints)
Tasks content/tasks/active.md
Journal content/tasks/journal.md
Meetings content/meetings/
Guide — (setup, how-to and FAQs; content lives in src/GuideView.tsx)
Claude sessions (panel on Day) reads ~/.claude/sessions + ~/.claude/projects — never written to

Deep-Dive Navigation

No child CLAUDE.md files yet. The places to look:

Path Description Keywords
README.md Setup, tokens, adding a section, gotchas install, setup, tokens, sections, tailwind
.claude/skills/ The skills that maintain this workspace skills, briefing, tasks, meetings
content/ Your markdown — the seeded files document their own formats tasks, journal, meetings, labels
content/memory/CLAUDE.md The wiki: durable knowledge, and the hub every page must be listed in memory, wiki, knowledge
.claude/scripts/setup/ One-time bootstrap: placeholders, isolated gws config, makefile setup, bootstrap, gws, auth
src/sections.ts The single list deciding which tabs you run sections, tabs, registry
src/GuideView.tsx The in-app guide — edit when you add or remove a section guide, setup, faq, docs
src/work-types.ts Your role's work-type buckets for the task board work types, role, buckets
server/index.ts Express server; mounts each section's API api, routes, content root
makefile Shortcuts: dev, build, run Claude, bootstrap, link check make, targets, setup, run
README.md#recommended-rtk rtk: compresses bash output before the agent reads it rtk, tokens, efficiency, cost

Skills

Skill Use it for
task-management Creating or editing any task in content/tasks/active.md
daily-briefing The morning picture: calendar, mail, PRs, backlog → today's Top 3
meeting-processor Pull yesterday's meetings from Granola into content/meetings/
sync-meetings The same, from a Google Drive folder instead
google-workspace-cli Driving the gws CLI (Drive, Docs, Sheets, Gmail, Calendar)
wiki Ingest documents into content/memory/, query it, lint it
memory-claude-md-sync Keeps the memory navigation honest on every wiki write
claude-md-template Keeps every CLAUDE.md to a consistent shape
/weekly-done-cleanup Prunes completed tasks out of the backlog into an archive
create-presentation Building a Fever-brand deck in Google Slides
/setup First-time setup: identity, placeholders, isolated Google auth
/wiki-ingest Ingesting one document into the wiki
/grill-me · /compact-session Interrogating a plan · capturing a session's context

Each skill needs setting up for you before first use — your GitHub login, Slack ID, Granola folders or Drive folder id. They say so at the top; fill in the blanks. Skills that depend on a tool you haven't authenticated skip that source and tell you, rather than inventing data.

Conventions

  • Tasks are atomic — one independently-checkable action per line, including in the Today Top 3. A shared meeting or theme doesn't justify bundling.
  • Today ↔ Standing are the same tickets. Build the Top 3 by selecting existing Standing items verbatim, not by writing a summary of them. Closing one closes its twin in the same edit.
  • The Top 3 is deliverables, never meetings. If a meeting matters, the task is its prep.
  • Every open task carries a priority — [P0] now, [P1] this cycle, [P2] later, [P3] nice-to-have (real work, but it never blocks anything). An untriaged task sorts above [P3], so leaving it off isn't a way to hide it.
  • Work type is the second marker — [Ship], [Quality], … from src/work-types.ts, which /setup generated for your role. It answers "what kind of work is this?", independently of urgency. Optional, but a board where most tasks have one is the board that can tell you where your week went.
  • External trackers own their issues. If a task lives in Jira, Linear or GitHub Issues, write through that tracker's API — never edit a local mirror.
  • Never put credentials or PII in content/. Reference a message or ticket id instead. These files are git-tracked.
  • Keep tool output small. Every oversized result is re-read on every later turn, so it costs repeatedly. Pipe Bash through grep/head rather than cat-ing whole files, use ranged Reads, and delegate broad searches to a subagent so the bulk stays out of the main context. rtk automates most of this for Bash — see the README. Note it does not cover Read/Grep/Glob, which is where the discipline still has to be yours.

Where your data lives

Path What Shared with anyone?
content/ Your tasks, journal, meeting notes, wiki No
content/sources/ Documents queued for wiki ingest (deleted once processed) No
.data/ Section runtime state No — gitignored
.data.example/ Example project + checkpoint, so the PR tab has something to show Committed, and it's fake
.env Your tokens No — gitignored

The installed @asucregonzalez/* packages are code only. They read the paths this app hands them and talk to no third party beyond the APIs you give tokens for.