Commands

The commands

In the loop: 2. Think deeper and clearer and 3. Decide. The tools you run by hand.

brainless has two kinds of command. The slash commands are the thinking tools: fifteen Markdown prompts in .wiki/_commands/, run by an agent (Claude Code or Gemini CLI) from the vault root. The wrappers in .claude/commands/ and .gemini/commands/ only point back to those files, so editing one prompt changes both agents. The shell commands are the plumbing: brainless <command> dispatches to a Python tool, no model in the loop unless the tool itself calls one. Two of them, dialectic and calibrate, exist in both forms and run the same logic.

Three rules apply to all of them, from .wiki/_commands/_shared-rules.md:

• They propose. You paste. Human folders (Work/, Personal/, Library/, Thinking/, Inbox/) are never written without an explicit "apply". The one exception is /calibrate, where your one-line answer is the approval.
• They quote, they do not paraphrase. A claim about a note comes with the exact line and the path.
• They file their output. Every run lands in .wiki/digests/queries/, so the next command can build on it. A run that is not filed is invisible.

Deciding

Command	What it asks	What it guards against
/decide "<question>"	Reversibility first (two-way door: decide today, three lines; one-way door: full note). Then criteria, beliefs quoted for and against each option, a forced third option, second-order effects, what would change your mind, a reference class and base rate before any confidence, and one dated action	Deliberating the reversible; criteria fitted to a favourite; confidence without a base rate; a decision with no action attached
/dialectic [topic]	Six personas, two rounds. Round one blind, round two answering the strongest objection. Scorecard computed by script, then a synthesis with a counterargument, what must be true, a cheap dated test and a probability. Empty topic argues today's captures	The first voice framing the rest; a synthesis written from mood; six agreeing voices mistaken for confirmation
/contradict [scope]	Belief against belief, belief against decision, belief against what you actually did in the last 14 days, with quotes from both sides. Lists beliefs unused for 90 days and proposes implicit beliefs you keep acting on	Drift between what you say you believe and what you do; a belief file that is a museum
/calibrate [decision]	Runs tools/calibrate.py, then one question per decision that is due: did the prediction happen? Records outcome and lesson beside the confidence. Asks for a base rate, a prediction and a review date on any decision that lacks them	Ungraded decisions; grading the result instead of the process; a confidence that was never a number

Thinking

Command	What it asks	What it guards against
/ideas [focus]	Three atomic ideas, each the intersection of at least two sources (a belief and a project, a resource and an area), each phrased as a question, each a paste-ready seed with a permanent id	Single-source ideas that are restatements; conclusions dressed as ideas
/pollinate <note>	Takes one new note, finds the beliefs it reinforces and the ones it challenges, checks which seeds it validates or kills, proposes two mutations	Reading only for agreement; new material that never touches old beliefs
/trace <topic>	Everything you wrote on a topic, oldest first, with the exact sentences and dates	Memory rewriting the past to match the present
/connect <A> <B>	The shortest link path between two notes and two to four edges that should exist and do not	Islands in the graph; connections that only exist in your head
/weekly [days]	Three themes from the window, one blind spot (an area or project absent from it), promotion candidates, a proposed diff to CONTEXT.md	The week that felt busy and touched nothing that matters
/context	Current priorities and projects against what the last seven days actually mention; flags drift both ways and names the stalest project	A context file that describes last month

Tending the vault

Command	What it does
/graduate [days]	Seeds older than the cutoff, scored by inbound links and length: promote to growing or evergreen, or flag the lonely ones for archive
/backlog [n]	The notes the wiki keeps linking to that nobody has written, ranked by demand, with paste-ready stubs for the top three
/sync	Reads the last seven digests and proposes log entries for the project notes they mention
/lint	Broken links, orphans, stale summaries, missing frontmatter; the top three actions
/recompile <path>	Forces one source through the compiler again

Shell commands

brainless help prints this list. Each line is one Python tool in tools/.

Command	What it does
compile [--dry-run|--full-rebuild|--only <phase>]	Build .wiki/ from your notes: summaries, concept pages, project mirrors, index
search "<query>" [--k N] [--json]	Search the compiled wiki
dialectic "<thesis>" [--run noon|evening|night] [--buzz] [--parallel]	Six personas argue it. Local by default, --buzz sends it through the live agents on the relay; one persona at a time by default, --parallel mentions all six at once; --scorecard prints the rolling 30 day scorecard
lint [--fix] [--fix-links] [--dry-run]	Wiki integrity checks and repairs, graph health, merge proposals, pages where the model talked about itself (--fix queues them for recompile)
calibrate	Decisions due for grading, decisions missing a prediction or a review date
today [--build|--send]	Preview, save or send the three-item Today queue; --action <id> ... answers an item
closeout [--dry-run]	Evening close-out: proposes one seed, one decision, one contradiction
digest	Nightly digest of Thinking/Daily/, archives the raw captures, compiles, lints
dashboard	Rebuild the projects view from every notes.md
health	Refresh and print _Agent-Context/HEALTH.md
file <command> "<title>" < note.md	File a result into .wiki/digests/queries/ so the wiki compounds
media add <url> | run | list	Queue a YouTube or Apple Podcasts episode link, work the transcript queue, show it
backup create | verify --identity <key> | status	Monthly age-encrypted archive of the vault and its history, the restore test, the dates of both
graph [--no-summaries] [--main-only] [--format graphml]	The link graph as a file Gephi opens, coloured by page type, into logs/graph/
eval [--min-hit5 N]	Ask the golden questions in _Agent-Context/retrieval-golden.json and score the search
chats triage|import|promote <file>	Bring a Claude or ChatGPT export in: list, import filtered, promote the keepers
export --out <dir> [--update]	Produce the public engine tree from a private vault, then leak-scan it
update	git pull and refresh dependencies
vault	Print the vault path
version	Print the engine version

The forms behind them

The commands are only as good as the templates they fill. _Templates/Decision.md carries the criteria-before-options rule, the third-option slot, the default_risk field (ego, emotion, social, inertia), the prediction block with confidence and review date, and an Outcome section that starts as "Pending review" and nags until it is not. _Templates/Belief.md carries "What Would Change My Mind" and a last_challenged date. _Templates/Idea.md carries a permanent zk: id and a rule that every idea links to at least one other note.

Why each of these exists is on Thinking Clearer.