projectmentor/hive-mind/wire-up

wire-up — self-provision tools & agents from cells

Hive Mind models each wireable thing as a cell (a kind:tool recipe or a kind:agent foreign-config integration). hv wire resolves a cell and provisions it: for tools it ensures the required credential, runs the platform-correct steps, and runs the cell's verify; for agents it reconciles the foreign config (e.g. Claude Code hooks). A comb is a named collection of cells.

Commands

hv wire <name>                 # provision a cell (auto-dispatch by its kind)
hv wire --comb <name>          # provision every cell in a comb
hv wire --list [--kind tool|agent]   # list cells (and combs)
hv wire --show <name>          # print a cell/comb definition (recipe + gotchas + verify)
hv wire --add <file.json>      # publish a cell/comb to the hive (syncs to your other devices)
hv wire <name> --env-file PATH # read tool credentials from a dotenv file (default ~/.claude/.env)

How to use it

1. See what exists / read the recipe first: hv wire --list, then hv wire --show <name> to

read its obtain instructions, gotchas, and what credential it requires.

1. Make the credential available — never via chat. If a tool requires a secret (e.g.

CLOUDFLARE_API_TOKEN, LOFTY_API_TOKEN), the user adds it to ~/.claude/.env themselves (e.g. tell them to run ! echo 'NAME=...' >> ~/.claude/.env). Do not ask them to paste a secret into the conversation. Better still, have them seal it into an encrypted capsule (hv capsule put <NAME> --env-file ~/.claude/.env --name <VAR>): hv wire reads an opened capsule first and falls back to the .env file. Then run hv wire <name> — the value flows file or capsule → tool, never the transcript.

1. Provision + verify: hv wire <name>. It is idempotent — if verify already passes it does

nothing. A clear message tells you what was missing, what ran, and whether verify passed.

1. Author a recipe: write a cell JSON (`{name, kind:"tool", requires:[…], spec:{obtain,

platforms:{linux|wsl|darwin:{steps:[…]}}, gotchas}, verify:{http|cmd, headers?, expect}}) and hv wire --add it.json. Built-in cells (like the claude` agent) resolve before journaled ones.

Notes

• kind:agent cells reconcile foreign config (Claude Code hooks today); hv wire claude replaced

the old hv doctor wire-agent, which 2.0 removed.

• If a tool's requires credential is missing, hv wire tells you exactly which one and where to

put it — surface that to the user rather than guessing.

• Real cells come from the hive (hv wire --list): they can carry private infrastructure details,

so they are published to the journal with hv wire --add, never committed. The repo's cells/ holds only a dummy example-tool.json to copy from.