robot_lab-am 0.0.1

checksums.yaml

@@ -0,0 +1,7 @@
+---
+SHA256:
+  metadata.gz: cd50cc952ad3b357990770041acd1bcce2dd0de11ecba0074e5affb1d167f555
+  data.tar.gz: 5e60e4ecbd1eff84b5998d63fea8f64387f7f546f7e666268481877ccf055626
+SHA512:
+  metadata.gz: 3e31eedae9a07189737093c35be19575d42a3307ef7523ed2221de5f08de0244144e33bdb1f8481875a360cfa6ac71f1ef4f07a6529f1978df59f7ade928def7
+  data.tar.gz: b512f414d2f4b1538ee11543dfdcbf64a6a4c161096feef9968b4e0359f33d56772a24da549374371e1de1fc2fb4f5aa29fc1b56365f584cc17ccc31cfc26c4c

data/.envrc

@@ -0,0 +1,6 @@
+# robot_lab_project/robot_lab-am/.envrc
+
+source_up
+
+export RR=`pwd`
+export BUNDLE_GEMFILE=Gemfile

data/.github/workflows/deploy-github-pages.yml

@@ -0,0 +1,52 @@
+name: Deploy Documentation to GitHub Pages
+on:
+  push:
+    branches:
+      - main
+      - develop
+    paths:
+      - "docs/**"
+      - "mkdocs.yml"
+      - ".github/workflows/deploy-github-pages.yml"
+  workflow_dispatch:
+
+permissions:
+  contents: write
+  pages: write
+  id-token: write
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      - name: Setup Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: 3.x
+
+      - name: Install dependencies
+        run: |
+          pip install mkdocs
+          pip install mkdocs-material
+          pip install mkdocs-macros-plugin
+          pip install mike
+
+      - name: Configure Git
+        run: |
+          git config --local user.email "action@github.com"
+          git config --local user.name "GitHub Action"
+
+      - name: Build MkDocs site
+        run: mkdocs build
+
+      - name: Deploy to GitHub Pages
+        uses: peaceiris/actions-gh-pages@v4
+        with:
+          github_token: ${{ secrets.GITHUB_TOKEN }}
+          publish_dir: ./site
+          keep_files: true

data/.loki

@@ -0,0 +1,10 @@
+# frozen_string_literal: true
+# robot_lab-am — watches repo activity (git, terminal, Claude Code sessions) to infer intent.
+
+import_up "repo_dev.loki"
+
+class Tasks
+  @@gem_name ||= "robot_lab-am".freeze
+
+  header "robot_lab-am v#{gem_version} — activity monitor for RobotLab"
+end

data/.rubocop.yml

@@ -0,0 +1 @@
+inherit_from: ../.rubocop-base.yml

data/Archspec.rb

@@ -0,0 +1,58 @@
+# robot_lab-am is a plain Ruby gem (RobotLab::Am), not a Rails app -- there is
+# no app/ tree, no controllers/models/views, so the :rails preset doesn't
+# apply. These are the actual boundaries described in ARCHITECTURE.md and
+# CLAUDE.md.
+
+component :watchers,      in: "lib/robot_lab/am/watchers/**/*.rb"
+component :event_log,     in: "lib/robot_lab/am/event_log.rb"
+component :inferrer,      in: "lib/robot_lab/am/inferrer.rb"
+component :intent_writer, in: "lib/robot_lab/am/intent_writer.rb"
+component :cli,           in: "lib/robot_lab/am/cli.rb"
+
+# NOTE: every file here reopens `module RobotLab` (shared with the external
+# robot_lab gem's own top-level module -- Inferrer calls RobotLab.build), so
+# a `dependencies.forbid` rule keyed on components would treat that one bare
+# `RobotLab.build` call as "depends on every component" -- every file
+# "defines" the bare RobotLab constant. Naming the actual target constants
+# instead of components sidesteps that ambiguity (same fix robot_lab's own
+# Archspec.rb applies for the same reason).
+
+# Watchers are pure activity collectors -- git/Claude Code/terminal in,
+# Event out (see ARCHITECTURE.md "Ingestion & normalization"). They must not
+# know how an Event gets stored, summarized, or reported; CLI is the only
+# thing that wires collection to storage/inference.
+watchers.cannot_reference_constants "RobotLab::Am::EventLog", "RobotLab::Am::Inferrer",
+                                    "RobotLab::Am::IntentWriter", "RobotLab::Am::CLI",
+                                    because: "a watcher's only job is producing Events -- what " \
+                                             "happens to an Event afterward is not its concern"
+
+# Inferrer consumes a bounded window of Events (already collected) and
+# produces an Intent; it must not reach back into how those Events were
+# gathered, or which specific source produced them.
+inferrer.cannot_reference_constants "RobotLab::Am::Watchers::GitWatcher",
+                                    "RobotLab::Am::Watchers::ClaudeWatcher",
+                                    "RobotLab::Am::Watchers::TerminalWatcher",
+                                    "RobotLab::Am::EventLog", "RobotLab::Am::IntentWriter",
+                                    "RobotLab::Am::CLI",
+                                    because: "Inferrer only knows the normalized Event shape, never " \
+                                             "a specific watcher, the log file, or how its own " \
+                                             "output is used"
+
+# IntentWriter is a pure renderer: Intent in, front-matter + prose file out.
+# It must not know how that Intent was produced.
+intent_writer.cannot_reference_constants "RobotLab::Am::Watchers::GitWatcher",
+                                         "RobotLab::Am::Watchers::ClaudeWatcher",
+                                         "RobotLab::Am::Watchers::TerminalWatcher",
+                                         "RobotLab::Am::EventLog", "RobotLab::Am::Inferrer",
+                                         "RobotLab::Am::CLI",
+                                         because: "IntentWriter only serializes an already-built " \
+                                                  "Intent -- it has no business asking how one was " \
+                                                  "inferred"
+
+# CLAUDE.md/ARCHITECTURE.md: GitWatcher subprocesses `git` and must never
+# build a command via string interpolation (the same rule robot_lab-to's
+# CommitManager enforces for its own git ops) -- Open3.capture3 with an
+# explicit argv array only.
+watchers.cannot_call :system, receiver: :none,
+                     because: "git subprocess calls must go through Open3.capture3 with an " \
+                              "argv array, never system()/backticks with an interpolated string"

data/CHANGELOG.md

@@ -0,0 +1,37 @@
+## [Unreleased]
+
+- **Standalone: dropped the robot_lab dependency.** Inference is a one-shot
+  prompt, so `Inferrer` now calls RubyLLM directly (scoped
+  `RubyLLM.context`, `assume_model_exists: true`) instead of building a
+  `RobotLab::Robot`. Dependencies are now just `ruby_llm` + `myway_config`,
+  making the gem useful outside the robot_lab-to environment. New `api_key`
+  setting / `RLAM_API_KEY` (cascades to the provider's conventional
+  env var, then a placeholder for keyless local servers); the
+  `ROBOT_LAB_RUBY_LLM__OPENAI_API_BASE` coupling is gone. Injectable seam
+  renamed `robot:` -> `chat:`.
+
+- Continuous daemon: `am start|stop|status` — forked/detached process with pid
+  file, per-tick heartbeat (`.robot_lab_am/heartbeat.json`), and debounced
+  inference (new-activity flag + `--debounce` window; failed inference warns
+  and retries instead of killing the daemon)
+- launchd integration: `am install|uninstall` write/remove a per-repo agent
+  plist (`RunAtLoad` + `KeepAlive`) that supervises `am start --foreground`
+- `Collector`: fingerprint-based dedupe against `events.jsonl`, so repeated
+  snapshots and daemon restarts never duplicate events
+- `Redactor`: masks credential-shaped values (key/token/secret/password
+  assignments, bearer tokens, known token formats) before events are stored
+  or sent to the LLM
+- `GitWatcher` no longer reports the gem's own `.robot_lab_am/` state
+  directory as uncommitted work
+- `am snapshot` now reports only *new* events and infers over the accumulated
+  recent window of the event log
+- MkDocs documentation site (Material theme) with GitHub Pages deploy workflow
+- myway_config-based configuration (`RobotLab::Am::Config`, same pattern as
+  robot_lab-to): bundled defaults → `~/.config/robot_lab_am/robot_lab_am.yml` →
+  `RLAM_*` env vars → CLI flags. Tunables: provider, model, api_base,
+  interval, debounce, inference_window, terminal_log. `am --help` now shows
+  the effective interval/debounce
+
+## [0.1.0] - 2026-08-31
+
+- Initial release

data/CLAUDE.md

@@ -0,0 +1,99 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## What This Gem Does
+
+`robot_lab-am` ("activity monitor") is a **standalone** gem (RobotLab-family
+naming, but no robot_lab dependency — just `ruby_llm` + `myway_config`) that
+watches a repo working directory — git activity, Claude Code session
+transcripts, and terminal commands — and infers the current goal or
+direction of work via a one-shot local-LLM call. The inferred intent seeds
+[`robot_lab-to`](https://github.com/MadBomber/robot_lab-to)'s takeover runs with
+real context instead of a cold objective string, but any tool can consume it.
+
+**Status: complete v1 — one-shot pipeline and continuous daemon.**
+`am snapshot` collects git/Claude Code/terminal activity, infers a goal via
+a local LLM, and writes `.robot_lab_am/current_intent.md`. `am start`/`stop`/
+`status` run the continuous daemon (detached, pid file + heartbeat,
+debounced inference), and `am install`/`uninstall` manage a launchd agent
+that supervises it. All verified end to end against a live daemon and a
+real LM Studio model. See `ARCHITECTURE.md` for the component survey and
+decisions.
+
+## Commands
+
+```bash
+bundle exec rake test          # all tests
+bundle exec rake test_file[path]  # single test file
+asgard quality                 # all *_check gates in parallel (tests + coverage,
+                               #   rubocop, flog, flay, reek, fasterer, typos, ...)
+asgard doc_builder             # build the MkDocs site
+asgard doc_server              # serve docs locally
+bin/console                    # IRB shell with gem loaded
+bin/am --help                  # CLI help
+bin/am snapshot [--repo PATH]  # one-shot: collect + infer + write intent
+bin/am start [--foreground]    # start the daemon (detached by default)
+bin/am stop|status             # stop / inspect the daemon
+bin/am install|uninstall       # manage the launchd agent plist
+```
+
+## Decided so far (see ARCHITECTURE.md for detail)
+
+- **Deployment model**: continuous background daemon, not an on-demand
+  snapshot — it should notice drift across sessions, not just seed one
+  takeover run.
+- **Terminal capture**: `~/.bashrc_history` is a single global `HISTFILE`
+  shared across every terminal and repo, with no timestamps or cwd, so it
+  can't attribute a command to a repo or a point in time. Instead,
+  `~/.bashrc__activity_monitor` (outside this repo, in the user's dotfiles)
+  registers a `preexec_functions` hook that appends
+  `epoch.microseconds\tcwd\tcommand` to `~/.activity_monitor/terminal_activity.log`
+  on every command. It's tab-delimited, not JSON, because it runs in
+  `preexec` — *before* the command executes — so it must not fork a
+  subprocess or every command gets delayed. Parsing/structuring that file
+  is this gem's job on the read side.
+- **Storage**: JSONL (`.robot_lab_am/events.jsonl` in the watched repo),
+  not SQLite — the only consumer is an LLM summarization pass, not ad
+  hoc queries.
+- **Inference model**: local, not hosted. `Inferrer` defaults to
+  `provider: :openai` / `model: "qwen/qwen3.8-27b"` against LM Studio's
+  OpenAI-compatible server (`lms server start`, `localhost:1234/v1`) —
+  explicitly not Anthropic. Your own activity log shouldn't have to
+  leave the machine, or need an API key, just to be summarized.
+- **RubyLLM directly, no robots** (2026-09-02): the inference is a
+  one-shot prompt, so `Inferrer` calls
+  `RubyLLM.context.chat(model:, provider:, assume_model_exists: true)`
+  in a scoped context (host apps' global RubyLLM config inherited,
+  never mutated). The gem no longer depends on `robot_lab` — deps are
+  `ruby_llm` + `myway_config` only, so it's usable outside the
+  robot_lab-to environment. Injectable seam is `chat:` (responds to
+  `#ask(prompt)` returning a message with `#content`).
+- **Configuration**: `Config < MywayConfig::Base` (same pattern as
+  robot_lab-to). Cascade: `config/defaults.yml` →
+  `~/.config/robot_lab_am/robot_lab_am.yml` (flat keys) →
+  `RLAM_*` env vars → CLI flags / constructor keywords.
+  Settings: provider, model, api_base, interval, debounce,
+  inference_window, terminal_log. `Am.config` is the memoized
+  process-wide instance (`Am.reset_config!` in tests). Caveat: the
+  defaults.yml-backed ivars are assigned by `super()` — never pre-nil
+  them in `Config#initialize`.
+
+- **Inference cadence**: debounced — the daemon polls watchers every
+  `--interval` seconds (default 15) but re-infers only when new events
+  have arrived since the last inference, and at most once per
+  `--debounce` seconds (default 300). A failed inference (LLM down) is
+  retried on the same debounce, and never kills the daemon.
+- **Redaction**: `Redactor` masks credential-shaped values (key/token/
+  secret/password assignments, bearer tokens, well-known token formats)
+  before any event is stored or sent to the LLM.
+- **Dedupe**: `Collector` fingerprints events against the existing
+  `events.jsonl`, so repeated snapshots/daemon restarts never duplicate
+  log lines. `wip` events dedupe on summary (their timestamps are
+  collection-time), everything else on timestamp + summary.
+
+## Testing
+
+Minitest with SimpleCov (branch coverage tracked). Same conventions as the
+other gems in this family — see the workspace-level `CLAUDE.md` and
+`.claude/rules/rubocop.md`.

data/LICENSE.txt

@@ -0,0 +1,21 @@
+The MIT License (MIT)
+
+Copyright (c) 2026 Dewayne VanHoozer
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.

data/README.md

@@ -0,0 +1,118 @@
+# robot_lab-am
+
+A standalone activity monitor that watches what's happening in a repo working
+directory — git activity, Claude Code session transcripts, and terminal
+commands — and infers the current goal/direction using a local LLM via
+[RubyLLM](https://rubyllm.com). The inferred intent is written to a well-known
+artifact (`.robot_lab_am/current_intent.md`) that any tool can read;
+[`robot_lab-to`](https://github.com/MadBomber/robot_lab-to) uses it to seed a
+takeover run with real context instead of a cold objective string, but nothing
+here depends on the [RobotLab](https://github.com/MadBomber/robot_lab)
+framework.
+
+**Documentation: [madbomber.github.io/robot_lab-am](https://madbomber.github.io/robot_lab-am)**
+
+## Installation
+
+Add to your Gemfile:
+
+```ruby
+gem "robot_lab-am"
+```
+
+Dependencies are just `ruby_llm` and `myway_config` — no robot framework
+required.
+
+## Usage
+
+```bash
+am snapshot                 # one-shot: collect activity, infer the goal, write
+                            # .robot_lab_am/current_intent.md in the repo
+
+am start                    # start the continuous daemon for this repo (detached)
+am status                   # is it running? heartbeat, event count, last inference
+am stop                     # stop it
+
+am install                  # write a launchd agent so the daemon runs at login
+am uninstall                # remove the launchd agent
+
+# All commands accept --repo PATH (default: current directory).
+# `am start` also accepts --foreground, --interval N (poll seconds, default 15),
+# and --debounce N (minimum seconds between inference runs, default 300).
+```
+
+The daemon polls three signal sources — git commits and uncommitted changes,
+your own messages in Claude Code session transcripts, and terminal commands run
+inside the repo — appends new (redacted, deduplicated) events to
+`.robot_lab_am/events.jsonl`, and re-infers `.robot_lab_am/current_intent.md`
+on a debounced cadence whenever activity arrives.
+
+### Prerequisites
+
+- **Local LLM**: inference defaults to LM Studio's OpenAI-compatible server on
+  `localhost:1234` (`lms server start`) so your activity log never leaves the
+  machine. Any RubyLLM provider/model can be passed to `Inferrer` instead.
+- **Terminal capture** (optional): commands are read from
+  `~/.activity_monitor/terminal_activity.log`, written by a bash `preexec` hook
+  (`~/.bashrc__activity_monitor` in your dotfiles). Without the hook, git and
+  Claude Code signals still work.
+
+## Configuration
+
+Settings cascade via [myway_config](https://github.com/MadBomber/myway_config),
+lowest to highest precedence:
+
+1. Bundled defaults
+2. User config file — `~/.config/robot_lab_am/robot_lab_am.yml` (flat keys)
+3. `RLAM_*` environment variables
+4. CLI flags (`--interval`, `--debounce`)
+
+### Environment variables
+
+| Variable | Default | Meaning |
+|----------|---------|---------|
+| `RLAM_PROVIDER` | `openai` | RubyLLM provider used for inference |
+| `RLAM_MODEL` | `qwen/qwen3.8-27b` | Model name |
+| `RLAM_API_BASE` | `http://localhost:1234/v1` | OpenAI-compatible server URL (applied to the `openai` provider) |
+| `RLAM_API_KEY` | *(none)* | API key; unset falls back to the provider's own env var, then a placeholder (fine for LM Studio) |
+| `RLAM_INTERVAL` | `15` | Seconds between daemon watcher polls |
+| `RLAM_DEBOUNCE` | `300` | Minimum seconds between inference runs |
+| `RLAM_INFERENCE_WINDOW` | `100` | Most recent events sent to the model |
+| `RLAM_TERMINAL_LOG` | `~/.activity_monitor/terminal_activity.log` | Preexec command log path |
+
+```bash
+# point inference at a different local model/server
+export RLAM_PROVIDER=openai
+export RLAM_MODEL=llama-3.3-70b
+export RLAM_API_BASE=http://localhost:8080/v1
+```
+
+Notes:
+
+- `am --help` prints the *currently effective* interval/debounce, so you can
+  see what your config file and environment resolve to.
+- API keys cascade: `RLAM_API_KEY` if set, else the chosen provider's
+  conventional env var (`OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, …), else a
+  placeholder — which is all a local LM Studio server needs. The default
+  local setup requires no key; your activity log never leaves the machine.
+- Inference runs in a scoped RubyLLM context, so embedding robot_lab-am in a
+  larger app never mutates that app's global RubyLLM configuration.
+
+## Development
+
+After checking out the repo, run `bin/setup` to install dependencies. Then run
+`bundle exec rake test` to run the tests. `bin/console` gives an interactive
+prompt with the gem loaded.
+
+```bash
+bundle exec rake test          # all tests
+asgard quality                 # all quality gates (tests + coverage, rubocop,
+                               #   flog, flay, reek, fasterer, typos, ...)
+asgard doc_builder             # build the MkDocs site
+asgard doc_server              # serve docs locally
+bin/console                    # IRB shell with gem loaded
+```
+
+## License
+
+The gem is available as open source under the terms of the [MIT License](https://opensource.org/licenses/MIT).

data/Rakefile

@@ -0,0 +1,30 @@
+# frozen_string_literal: true
+
+# Quality gates (quality, rubocop_check, flog_check, flay_check, ...),
+# documentation tasks (doc_builder, doc_server), and the gem lifecycle
+# (build, install, release) live in asgard — see .loki and the shared
+# dev/*.loki files it imports. This Rakefile keeps only the task asgard
+# itself delegates to: the test suite.
+
+require 'rake/testtask'
+
+Rake::TestTask.new(:test) do |t|
+  t.libs << 'test'
+  t.libs << 'lib'
+  t.test_files = FileList['test/**/*_test.rb', 'test/**/test_*.rb'].exclude('**/*_helper.rb')
+  t.verbose = true
+  t.ruby_opts << '-rtest_helper'
+end
+
+task default: :test
+
+desc 'Run tests with verbose output'
+task :test_verbose do
+  ENV['TESTOPTS'] = '--verbose'
+  Rake::Task[:test].invoke
+end
+
+desc 'Run a single test file'
+task :test_file, [:file] do |_t, args|
+  ruby "test/#{args[:file]}"
+end

data/bin/am

@@ -0,0 +1,13 @@
+#!/usr/bin/env ruby
+# frozen_string_literal: true
+
+# Activate bundler if running from the gem's own dev environment.
+if File.exist?(File.expand_path("../Gemfile", __dir__))
+  require "bundler/setup"
+end
+
+require "robot_lab/am"
+
+Signal.trap("INT") { exit 130 }
+
+RobotLab::Am::CLI.run(ARGV)

data/docs/assets/pipeline.svg

@@ -0,0 +1,100 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 900 700" font-family="Helvetica, Arial, sans-serif" role="img" aria-label="robot_lab-am pipeline">
+  <defs>
+    <marker id="a" markerWidth="9" markerHeight="9" refX="7" refY="3" orient="auto" markerUnits="strokeWidth">
+      <path d="M0,0 L7,3 L0,6 Z" fill="#94a3b8"/>
+    </marker>
+  </defs>
+  <style>
+    .t  { fill:#ffffff; font-size:13px; font-weight:600; }
+    .s  { fill:#e2e8f0; font-size:10px; }
+    .hd { fill:#cbd5e1; font-size:12px; font-weight:700; letter-spacing:.5px; }
+    .lg { fill:#cbd5e1; font-size:11px; }
+  </style>
+
+  <!-- signal sources -->
+  <text x="40" y="14" class="hd">SIGNAL SOURCES (human activity)</text>
+  <rect x="40" y="24" width="250" height="46" rx="10" fill="#64748b"/>
+  <text x="165" y="45" text-anchor="middle" class="t">Git repo</text>
+  <text x="165" y="61" text-anchor="middle" class="s">commits · uncommitted changes</text>
+
+  <rect x="325" y="24" width="250" height="46" rx="10" fill="#64748b"/>
+  <text x="450" y="45" text-anchor="middle" class="t">Claude Code transcripts</text>
+  <text x="450" y="61" text-anchor="middle" class="s">~/.claude/projects/&lt;slug&gt;/*.jsonl</text>
+
+  <rect x="610" y="24" width="250" height="46" rx="10" fill="#64748b"/>
+  <text x="735" y="45" text-anchor="middle" class="t">Terminal preexec log</text>
+  <text x="735" y="61" text-anchor="middle" class="s">~/.activity_monitor/terminal_activity.log</text>
+
+  <line x1="165" y1="70" x2="165" y2="98" stroke="#94a3b8" stroke-width="2" marker-end="url(#a)"/>
+  <line x1="450" y1="70" x2="450" y2="98" stroke="#94a3b8" stroke-width="2" marker-end="url(#a)"/>
+  <line x1="735" y1="70" x2="735" y2="98" stroke="#94a3b8" stroke-width="2" marker-end="url(#a)"/>
+
+  <!-- watchers -->
+  <rect x="40" y="100" width="250" height="46" rx="10" fill="#0891b2"/>
+  <text x="165" y="121" text-anchor="middle" class="t">GitWatcher</text>
+  <text x="165" y="137" text-anchor="middle" class="s">recent commits + WIP (own state dir filtered)</text>
+
+  <rect x="325" y="100" width="250" height="46" rx="10" fill="#0891b2"/>
+  <text x="450" y="121" text-anchor="middle" class="t">ClaudeWatcher</text>
+  <text x="450" y="137" text-anchor="middle" class="s">your own plain-language messages</text>
+
+  <rect x="610" y="100" width="250" height="46" rx="10" fill="#0891b2"/>
+  <text x="735" y="121" text-anchor="middle" class="t">TerminalWatcher</text>
+  <text x="735" y="137" text-anchor="middle" class="s">only commands with cwd inside the repo</text>
+
+  <line x1="165" y1="146" x2="435" y2="196" stroke="#94a3b8" stroke-width="2" marker-end="url(#a)"/>
+  <line x1="450" y1="146" x2="450" y2="196" stroke="#94a3b8" stroke-width="2" marker-end="url(#a)"/>
+  <line x1="735" y1="146" x2="465" y2="196" stroke="#94a3b8" stroke-width="2" marker-end="url(#a)"/>
+
+  <!-- collector -->
+  <rect x="300" y="198" width="300" height="52" rx="10" fill="#f59e0b"/>
+  <text x="450" y="220" text-anchor="middle" class="t">Collector</text>
+  <text x="450" y="237" text-anchor="middle" class="s">Redactor masks secrets · fingerprint dedupe</text>
+  <line x1="450" y1="250" x2="450" y2="278" stroke="#94a3b8" stroke-width="2" marker-end="url(#a)"/>
+
+  <!-- event store -->
+  <rect x="325" y="280" width="250" height="46" rx="10" fill="#475569"/>
+  <text x="450" y="301" text-anchor="middle" class="t">.robot_lab_am/events.jsonl</text>
+  <text x="450" y="317" text-anchor="middle" class="s">append-only event log, one JSON line per event</text>
+  <line x1="450" y1="326" x2="450" y2="354" stroke="#94a3b8" stroke-width="2" marker-end="url(#a)"/>
+
+  <!-- inferrer + local LLM -->
+  <rect x="325" y="356" width="250" height="52" rx="10" fill="#a855f7"/>
+  <text x="450" y="378" text-anchor="middle" class="t">Inferrer</text>
+  <text x="450" y="395" text-anchor="middle" class="s">bounded recent window (last 100 events)</text>
+
+  <rect x="640" y="356" width="220" height="52" rx="10" fill="#10b981"/>
+  <text x="750" y="378" text-anchor="middle" class="t">Local LLM</text>
+  <text x="750" y="395" text-anchor="middle" class="s">LM Studio · localhost:1234 · no API key</text>
+  <line x1="575" y1="376" x2="638" y2="376" stroke="#94a3b8" stroke-width="2" marker-end="url(#a)"/>
+  <line x1="638" y1="392" x2="577" y2="392" stroke="#94a3b8" stroke-width="2" marker-end="url(#a)"/>
+
+  <line x1="450" y1="408" x2="450" y2="436" stroke="#94a3b8" stroke-width="2" marker-end="url(#a)"/>
+
+  <!-- intent artifact -->
+  <rect x="325" y="438" width="250" height="52" rx="10" fill="#6366f1"/>
+  <text x="450" y="460" text-anchor="middle" class="t">.robot_lab_am/current_intent.md</text>
+  <text x="450" y="477" text-anchor="middle" class="s">goal · confidence · evidence · open questions</text>
+  <line x1="450" y1="490" x2="450" y2="518" stroke="#94a3b8" stroke-width="2" marker-end="url(#a)"/>
+
+  <!-- consumer -->
+  <rect x="325" y="520" width="250" height="46" rx="10" fill="#0ea5e9"/>
+  <text x="450" y="541" text-anchor="middle" class="t">robot_lab-to</text>
+  <text x="450" y="557" text-anchor="middle" class="s">seeds a takeover run with real context</text>
+
+  <!-- daemon loop -->
+  <rect x="40" y="280" width="220" height="128" rx="10" fill="none" stroke="#22d3ee" stroke-width="2" stroke-dasharray="6 4"/>
+  <text x="150" y="304" text-anchor="middle" class="t">Daemon (am start)</text>
+  <text x="150" y="326" text-anchor="middle" class="lg">polls watchers every 15s</text>
+  <text x="150" y="344" text-anchor="middle" class="lg">debounced inference (≥300s apart)</text>
+  <text x="150" y="362" text-anchor="middle" class="lg">heartbeat.json every tick</text>
+  <text x="150" y="380" text-anchor="middle" class="lg">pid file · SIGTERM stop</text>
+  <text x="150" y="398" text-anchor="middle" class="lg">supervised by launchd (am install)</text>
+  <line x1="260" y1="344" x2="298" y2="240" stroke="#22d3ee" stroke-width="2" stroke-dasharray="6 4" marker-end="url(#a)"/>
+  <line x1="260" y1="382" x2="323" y2="382" stroke="#22d3ee" stroke-width="2" stroke-dasharray="6 4" marker-end="url(#a)"/>
+
+  <!-- legend -->
+  <text x="40" y="620" class="hd">ONE-SHOT PATH</text>
+  <text x="40" y="640" class="lg">am snapshot runs the same pipeline once — collect → infer → write — with no daemon.</text>
+  <text x="40" y="660" class="lg">Everything runs on your machine: the activity log is only ever summarized by a local model.</text>
+</svg>