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. The inferred intent is written to a well-known artifact (.robot_lab_am/current_intent.md) that any tool can read; 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 framework.

Documentation: madbomber.github.io/robot_lab-am

Add to your Gemfile:

gem "robot_lab-am"

Dependencies are just ruby_llm and myway_config — no robot framework required.

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.

• 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.

Settings cascade via 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)

# 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.

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.

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

The gem is available as open source under the terms of the MIT License.