feat(cli): add brain hooks so the memory loop runs by itself

.brain/.gitignore

@@ -0,0 +1,2 @@
+# Machine-local lock files, recreated on demand.
+*.lock

.brain/brain.yml

@@ -0,0 +1,52 @@
+# brAIn configuration
+#
+# Every setting below is shown at its built-in default and commented out.
+# Uncomment a line to override it. Deleting this file is safe — brAIn runs
+# on the defaults with no config at all.
+#
+# Precedence, lowest to highest: built-in defaults, this file, environment
+# variables. Run "brain config" to see where each value is coming from.
+
+# Storage backend. Only "markdown" is available today; "pg" is planned.
+# Override with $BRAIN_STORAGE.
+# storage: markdown
+
+# PostgreSQL connection, used only when storage is "pg".
+# Override with $BRAIN_DATABASE_URL.
+# database:
+#   url: postgres://brain:secret@localhost:5432/brain_dev
+
+# Embedding model for semantic recall, used only when storage is "pg".
+# Providers: openai, anthropic, ollama.
+# Override with $BRAIN_EMBEDDING_PROVIDER and $BRAIN_EMBEDDING_MODEL.
+# embeddings:
+#   provider: openai
+#   model: text-embedding-3-small
+
+# Trust ladder: clean outcomes needed to climb each rung. The counter resets
+# to zero at every promotion, so these are per-level, not cumulative.
+#
+# These gate the edges BETWEEN levels, not the levels themselves. Raising a
+# threshold does not demote a domain that is already past it — a domain at
+# auto_ship stays there no matter what you set promote_to_notify to. Use
+# "brain trust override" to move a domain down deliberately.
+# trust:
+#   promote_to_notify: 10
+#   promote_to_auto_ship: 30
+#   promote_to_full_auto: 100
+#   # Total to reach full_auto: 140
+
+# How long a fact stays fresh. Stamped onto each fact when it is written,
+# so changing this affects new facts only — facts already on disk keep the
+# expiry they were created with.
+# facts:
+#   stale_after_days: 30
+
+# Clean-outcome streak at which a lesson retires. A lesson can override
+# this with its own retire_after frontmatter field.
+#
+# Retirement is permanent. Raising this does nothing for lessons the old,
+# lower value already retired. Lowering it can retire several lessons at once
+# on the next clean outcome.
+# lessons:
+#   retire_after_streak: 20

.brain/corrections/2026-09-18-no-ai-or-claude-attribution-in-commit-messages.md

@@ -0,0 +1,7 @@
+---
+layer: correction
+domain: workflow
+created: 2026-09-18T08:16:52.051511Z
+source: human
+---
+No AI or Claude attribution in commit messages

.brain/corrections/2026-09-18-no-placeholders-or-stubs-if-a-piece-is-unbuildable-surface-i.md

@@ -0,0 +1,7 @@
+---
+layer: correction
+domain: workflow
+created: 2026-09-18T08:16:52.054917Z
+source: human
+---
+No placeholders or stubs — if a piece is unbuildable, surface it in the pitch scope rather than faking it

.brain/decisions/brain-prints-editor-configuration-and-never-writes-it-mergin.md

@@ -0,0 +1,7 @@
+---
+layer: decision
+domain: cli
+created: 2026-09-18T08:16:43.927566Z
+source: human
+---
+brAIn prints editor configuration and never writes it; merging into a settings file the user owns is not a memory tool's job

.brain/decisions/commits-follow-conventional-commit-types-with-imperative-sub.md

@@ -0,0 +1,7 @@
+---
+layer: decision
+domain: workflow
+created: 2026-09-18T08:16:43.917673Z
+source: human
+---
+Commits follow conventional commit types with imperative subjects of 72 characters or less

.brain/decisions/configuration-resolves-built-in-defaults-first-then-brain-ym.md

@@ -0,0 +1,7 @@
+---
+layer: decision
+domain: config
+created: 2026-09-18T08:16:43.910867Z
+source: human
+---
+Configuration resolves built-in defaults first, then brain.yml, then BRAIN_* environment variables

.brain/decisions/design-docs-pitches-and-decisions-live-in-doc-which-is-gitig.md

@@ -0,0 +1,7 @@
+---
+layer: decision
+domain: workflow
+created: 2026-09-18T08:16:43.914258Z
+source: human
+---
+Design docs, pitches, and decisions live in .doc/, which is gitignored and stays out of the public repo

.brain/decisions/docs-only-changes-land-on-main-directly-anything-touching-co.md

@@ -0,0 +1,7 @@
+---
+layer: decision
+domain: workflow
+created: 2026-09-18T08:16:43.920987Z
+source: human
+---
+Docs-only changes land on main directly; anything touching code goes through a branch and a PR

.brain/decisions/effectiveness-files-carry-no-title-heading-the-display-title.md

@@ -0,0 +1,7 @@
+---
+layer: decision
+domain: storage
+created: 2026-09-18T08:16:43.93096Z
+source: human
+---
+Effectiveness files carry no title heading — the display title is derived from persona and domain so there is only one copy of it

.brain/decisions/go-compiled-to-a-single-static-binary-with-minimal-dependenc.md

@@ -0,0 +1,7 @@
+---
+layer: decision
+domain: architecture
+created: 2026-09-18T08:16:43.899489Z
+source: human
+---
+Go, compiled to a single static binary with minimal dependencies, so installing brAIn is downloading one file

.brain/decisions/markdown-files-are-the-default-store-postgresql-with-pgvecto.md

@@ -0,0 +1,7 @@
+---
+layer: decision
+domain: storage
+created: 2026-09-18T08:16:43.903653Z
+source: human
+---
+Markdown files are the default store; PostgreSQL with pgvector is an optional upgrade and never a requirement

.brain/decisions/trust-fact-and-lesson-thresholds-are-settable-only-in-brain.md

@@ -0,0 +1,7 @@
+---
+layer: decision
+domain: config
+created: 2026-09-18T08:16:43.907156Z
+source: human
+---
+Trust, fact, and lesson thresholds are settable only in brain.yml, never by environment variable — a threshold set by an invisible variable is a trust ladder nobody can audit

.brain/decisions/trust-promotion-is-evaluated-only-inside-record-on-a-clean-o.md

@@ -0,0 +1,7 @@
+---
+layer: decision
+domain: trust
+created: 2026-09-18T08:16:43.924238Z
+source: human
+---
+Trust promotion is evaluated only inside Record on a clean outcome — loading trust state never promotes, so reads never become locked writes

.brain/lessons/a-session-start-hook-that-exits-non-zero-blocks-the-session.md

@@ -0,0 +1,7 @@
+---
+layer: lesson
+domain: cli
+created: 2026-09-18T08:16:52.036925Z
+source: human
+---
+A session-start hook that exits non-zero blocks the session, and brain recall exits 2 on an empty store — every hook consumer must absorb both

.brain/lessons/definition-doc-examples-drift-from-the-cli-06-mcp-and-cli-do.md

@@ -0,0 +1,7 @@
+---
+layer: lesson
+domain: docs
+created: 2026-09-18T08:16:52.040788Z
+source: human
+---
+Definition-doc examples drift from the CLI: 06-mcp-and-cli documents brain recall --for council, a flag that never existed. Verify every command in a template against the real command tree

.brain/lessons/git-carries-files-not-directories-an-empty-layer-directory-v.md

@@ -0,0 +1,7 @@
+---
+layer: lesson
+domain: cli
+created: 2026-09-18T08:16:52.048041Z
+source: human
+---
+git carries files, not directories: an empty layer directory vanishes from a clone unless something occupies it

.brain/lessons/prescriptive-rules-read-like-corrections-but-behave-like-les.md

@@ -0,0 +1,7 @@
+---
+layer: lesson
+domain: engine
+created: 2026-09-18T08:16:52.044478Z
+source: human
+---
+Prescriptive rules read like corrections but behave like lessons — filing one as a correction makes it permanent and impossible to retire

README.md

@@ -8,7 +8,7 @@ Not a vector database. Not a knowledge graph. Not an AI memory product. The ligh
 
 ## Status
 
-Pre-alpha. The memory engine, CLI, MCP server, trust ladder, effectiveness tracking, and configuration are implemented. The PostgreSQL adapter and editor integrations are coming in future pitches.
+Pre-alpha. The memory engine, CLI, MCP server, trust ladder, effectiveness tracking, configuration, and editor integration are implemented. The PostgreSQL adapter is coming in a future pitch.
 
 ## Install
 
@@ -38,6 +38,34 @@ brain config                  # what's in effect, and where each value came from
 
 Commit `.brain/` to git so your memory travels with the project.
 
+## Editor Setup
+
+Everything above is something you have to remember to run, which is the one
+thing memory should not depend on. To make the loop run by itself:
+
+```bash
+brain hooks                   # Claude Code: settings.json block, then CLAUDE.md rules
+brain hooks --tool cursor     # Cursor: .cursor/rules/brain.mdc
+```
+
+brAIn prints; you paste. It never edits your settings, your `CLAUDE.md`, or
+anything else you own — merging into a file you have already customised is
+not a memory tool's job.
+
+In Claude Code the hook loads your highest-authority memories into every new
+session, and again after a `/clear` or a compaction. Compaction is the one
+that matters most: it is where a correction gets summarised away and the
+model quietly loses the thing it was told not to do.
+
+The rules cover what no hook can. A hook cannot know which domain you are
+touching, so per-domain recall, acting on the trust level, and restraint
+about what gets written down are all instructions to the model rather than
+automation.
+
+Memories are committed, so they travel with the repository. The hook
+configuration is per-developer: everyone who clones runs `brain hooks` once
+on their own machine.
+
 ## Configuration
 
 Everything works with no configuration. To change a default, uncomment the