mastermind-brain
v0.32.1
Published
A markdown brain that gives your AI coding tools judgment and rigor: sharp defaults, real decisions, and the discipline to verify before saying done. For Claude Code, Cursor and Codex. Versioned, provenance-signed installer.
Maintainers
Readme
mastermind-brain
Installs MasterMind, a markdown brain that gives your AI coding tools judgment and rigor: sharp defaults, real decisions, and the discipline to verify before saying done. For Claude Code, Cursor and Codex.
You don't learn commands. You describe what you want, and it applies the right discipline: 22 skills and 4 specialist agents it reaches for on its own, from debugging to review to keeping secrets out of git.
cd my-project
npx mastermind-brain # install into this project (all your AI tools)
npx mastermind-brain --global # Claude Code in every project
npx mastermind-brain check # doctor: is this project wired?
npx mastermind-brain update # refresh the brain + repair links
npx mastermind-brain uninstall # clean removalCommands an agent calls for itself
MasterMind's skills are not native to Cursor or Codex, so an agent there either pastes the whole
library into context or guesses a path. These lookups answer from whichever brain the current
directory belongs to: the project's own .mastermind/ first, then the shared clone:
mastermind skills # the routing table: every skill, one line each
mastermind skill performance # one skill's full instructions
mastermind agents # the isolated-context roles
mastermind route "why is this slow?" # the table again, with keyword matches arrowed
mastermind wrong-log # every time MasterMind was wrong, and what caught it
mastermind conflicts # what else is installed, and where it overlaps usAdd --json to any of them for structured output. They are read-only: no install, no
network, no writes: if no brain is installed they say so and exit non-zero. Without a global
install, call the shim in the clone: ~/.mastermind/bin/mastermind skills.
Most machines end up with several skill packs installed. conflicts lists the foreign skills it
can see and where their triggers overlap ours; it reports rather than resolves, because measured
routing barely moves in a crowded install (7/8 either way). Precedence when they disagree: your
project's own skills → installed packs → MasterMind defaults, and on a rule conflict the stricter
rule wins.
route deliberately does not shortlist. Keyword overlap picks the right skill for requests
phrased like the descriptions, but only hints correctly for 2 of 8 requests phrased the way
people actually talk ("the invoice screen takes nine seconds to open"), so it marks candidates
and always returns every option: hiding the right skill behind a confident guess would be
worse than not guessing. The model routes; this just puts the table in front of it.
Why npx over curl | bash: every release is a versioned, immutable, provenance-signed npm
artifact, and fresh installs pin the brain to the matching git tag: you always know exactly
what ran. The brain lives at ~/.mastermind (a plain git repo you can read end to end);
projects get their own committed copy in <project>/.mastermind/.
MIT · source
