keyshroud

Least privilege: a project's own names

By default keyshroud run puts every wired secret into the command, so an agent working on one project can use the tokens of all the others. A scope narrows that for a directory:

cd ~/code/shop
keyshroud scope --only 'STRIPE_*,GH_TOKEN'   # writes ./.keyshroud.json
keyshroud scope                              # what is in scope here, and how many names are not
{ "only": ["STRIPE_*", "GH_TOKEN"] }

In that directory and below it (the nearest .keyshroud.json going up decides):

  • keyshroud run, secrets_run, env, env --shell and pi/omp's shell get only the names in scope; run and secrets_run do not pass a wired name outside it on, not even from the environment the agent was started with.
  • run --only NAME, get NAME and secrets_run with only refuse a name outside the scope, and say so.
  • The names Claude Code is told at session start, and secrets_list, are the ones in scope.
  • Masking is not narrowed: a value outside the scope that turns up in output is still masked (a name wired to a team secret manager from the moment its value has been read there: masking never contacts a source by itself).
  • secrets_run takes the scope of the directory its command runs in (cwd; through a link, the directory it leads to), secrets_list the one of the directory the MCP server was started in.
  • add and wire say when the name they just wired is outside the scope, so you can add it there.

The file holds names and patterns only (* stands for any run of characters), so it can be committed; a teammate's machine applies it to whatever they have wired under those names. A file that cannot be read as a scope (also a link to a file that is gone) puts nothing in scope and says why, instead of falling back to everything or to a scope further up. References (--refs .env.refs, get keyshroud://Item) are not narrowed: they already name what they want. A scope is a default against accidents, not a wall: see the security model.

Edit this page on GitHub · Markdown