Commands

cognitiveOS ships five commands. The CLI is a scaffolding tool, not a runtime. init generates your filesystem once, and your agent plus hooks maintain state from then on.

cognitiveos init     # one-time setup (3 questions)
cognitiveos start    # Mission Control, where you left off, in your terminal
cognitiveos dump "thought"   # capture anything, zero friction
cognitiveos check    # verify everything is wired correctly
cognitiveos install-skill    # install the global /cognitiveos skill (once per machine)

init

One-time setup. A 3-question wizard generates your filesystem: the zone folders, STATE.md, and the agent routing files (AGENTS.md + CLAUDE.md).

npx cognitiveos init

Atomic. Generation either completes or makes no changes.
Never overwrites. Existing files are left untouched; init is additive.
Finishes in well under a minute.

start

Prints Mission Control to your terminal: a snapshot of where you left off, read straight from STATE.md. It leads with the ➡ PICK UP line from your Session Handoff (the exact next action), then current focus, blockers, and open loops. Renders in under half a second.

cognitiveos start

start also has a machine-only hook mode, used by the deterministic session hooks (see Hooks):

npx cognitiveos start --hook --agent=claude        # or --agent=antigravity

--hook reads the agent's hook JSON on stdin instead of printing to a terminal. On session start it emits the Mission Control injection envelope. On a Claude Code Stop event, if STATE.md hasn't been saved today it blocks with a one-time reminder to do the end-session update. Never throws, never breaks the host agent CLI.

Wire start (no flags) to run automatically on every directory change:

# .zshrc / .bashrc
cd() { builtin cd "$@" && cognitiveos start 2>/dev/null; }

# PowerShell $PROFILE
function cd { Set-Location @args; cognitiveos start 2>$null }

dump

Friction-free capture. Appends a thought to brain-dump/ without breaking your flow. It never fails and never asks follow-up questions.

cognitiveos dump "refactor the auth middleware before shipping"

Capture now, triage later. Getting it out of your head is the whole point.

check

Verifies your cognitiveOS install is wired correctly: a silent-failure detector that runs a set of integrity checks and reports drift.

cognitiveos check          # report problems
cognitiveos check --fix    # repair what it safely can

Checks cover: STATE.md (7 sections present, size warning past ~150 lines), first-run setup status, active-projects cap (max 3), inbox rot (10+ items or oldest ≥7 days waiting triage), handoff staleness, CLAUDE.md/AGENTS.md drift, all 6 zone CONTEXT.md files, the one-task invariant in focus/current-task.md, a Done when: stop condition on the current task, the 3 slash hooks, sessions/ writability, the shared loop block, and that the deterministic session hook (plus, for Claude Code, the Stop hook) is wired for each installed agent.

Most of these are soft ⚠ advisories, not failures: they surface drift and rot without blocking you. --fix auto-repairs what's safely mechanical: renames legacy memory.md → STATE.md, regenerates CLAUDE.md from AGENTS.md, restores any missing zone CONTEXT.md, and re-wires missing session hooks. It never touches STATE.md content or anything you wrote.

Run it whenever something feels off, or after editing the generated files by hand.

install-skill

Installs the distributable /cognitiveos skill into your home directory, once per machine, so any future project can invoke it directly without running init first.

cognitiveos install-skill                  # defaults to --agent claude
cognitiveos install-skill --agent codex
cognitiveos install-skill --agent antigravity
cognitiveos install-skill --agent all

Writes SKILL.md to ~/.claude/skills/cognitiveos/, ~/.codex/skills/cognitiveos/, and/or ~/.agents/skills/cognitiveos/ depending on --agent. If a copy already exists and differs, it's backed up to a timestamped .bak first, then replaced. Safe to re-run any time you upgrade the CLI. This is separate from the project skill init generates; see Agent Setup.

Next steps

The Zones: what each folder means.
Hooks: the automation that updates STATE.md for you.
The Keeper: the maintenance subagent and first-run interview.