Reference

Lookup material: exact flags, defaults, file paths, and exit codes. These pages answer "what exactly does this accept?" — for "how do I do X?", start with the operator guide.

Command line

Page	Covers
loop.sh	The Ralph outer loop: modes, every flag, environment variables, exit codes, stop conditions
swarm.sh	Parallel worktree streams: flags, partitioning rules, lane safety, telemetry output
Installers	install.sh and install.ps1: flags, targets, environment variables, verification

Validation and files

Page	Covers
Gates	Every check and harness this repository ships, what each proves, and when it runs
Artifacts	Every file wgm reads and writes, where it puts them, and what to gitignore

Quick answers

Question	Answer
How do I validate a change to this repo?	make validate
How do I stop a running loop?	Ctrl+C, or touch .wgm/STOP
Where does telemetry land?	.wgm/metrics.tsv, and .wgm/metrics/ for swarm lanes
Why won't build start?	It requires an IMPLEMENTATION_PLAN.md in the root or .wgm/
Which agent does the loop call?	$WGM_AGENT, --agent "CMD", or argv after --
How do I skip the companion skills?	--no-companions (bash) or -NoCompanions (PowerShell)
Where do I turn off upstream reporting?	.github/wgm-hive.yml — set consent: false

Deeper protocol references

The references/ directory holds the terse, load-every-iteration rules the agent itself reads. They are denser than these pages and written for the agent, not the operator:

grilling · ralph-loop · telemetry · scenarios · scoring · subagents · artifacts · stall-recovery · hard-to-test-domains · validation-env · self-improvement · heuristics