The entry point

ssh in, type claude. The human says it; Claude runs the be CLI.

There is one way in. You ssh to the dev workstation, you are in your home directory, you type claude. The harness is already there.

  ssh <you>@dev-box
        │
        ▼
  ~/  ── CLAUDE.md loaded, rules in effect
      ── SessionStart hook prints the map: every be-* here, what is open
      ── advisor attached
        │
        ▼
  claude>  "I keep thinking about how we'd ingest zoom transcripts"
        │
        ▼
  Claude runs:   be wiki new zoom-etl
  Claude says:   cd ~/be-wiki-zoom-etl && claude    then /dump

No setup command, no briefing step. The entry point is the setup.

You do not type the CLI

You can — be help is a normal shell command — but the design assumes you won’t. Claude has be pre-approved in ~/.claude/settings.json and turns what you say into what it runs:

  "idea about X"            be wiki new <name>
  "look at Sam's wiki"      be wiki pull <name>
  "what's in flight"        be wiki list
  "is this box ok"          be doctor

The CLI is built for that caller: it never prompts, it fails fast with a message, and its output is one item per line. See The CLI.

Home, then a repo

~ is where you start things and find things. Work happens inside a repo, where that repo’s own harness takes over:

  ~                              ~/be-wiki-zoom-etl
  /gm   the map                  /gm   this wiki: pages, open questions
  /new  be <type> new            /dump /interview /brief /graduate
  /pull be <type> pull           writer · reader agents
  /gn   land every repo          /gn   land this one

Claude cannot change your shell’s directory. It tells you the cd line; you type it (or eval "$(be shell-init)" in your rc lets be … go do it for you).

What is in scope

The team is the same everywhere. What changes is the folder.