Operator guide

How to install, run, and validate wgm. This is the human-facing half of the docs — the agent/ half is for the agent following the protocol. Start here, then jump to the detail page you need.

Executive overview

• Who this is for: the human operating wgm — installing it, driving the build loop, and troubleshooting — not the agent that follows the protocol.
• What wgm is: an Agent Skill (a SKILL.md folder) that turns a rough request into working software through a grilled plan and a test-validated build loop.
• The operator journey: install once → invoke /wgm (or drive scripts/loop.sh) → reach for a container only when a scenario needs a live service → consult troubleshooting when a gate fails.
• The one mental model: wgm steers on backpressure — a deterministic pass/fail check. Your job is to sit on the loop (add validation checks, set thresholds), not in it.
• Least you need to start: run the installer, then invoke /wgm with your request. Everything below is optional depth.

The operator journey

flowchart LR
  I[Install wgm] --> U["Invoke /wgm in-session"]
  U --> Q{Large or ambiguous build?}
  Q -- yes --> L["Drive scripts/loop.sh: fresh context each pass"]
  Q -- no --> S[Ship / Handoff]
  L --> C{Scenario needs a live service?}
  C -- yes --> K[Validate in an OCI container]
  C -- no --> S
  K --> S
  L -. a gate fails .-> T[Troubleshooting]
  I -. wgm doesn't appear .-> T

Where to go next

New to wgm? Start with Get started — it covers install through your first validated build in eight steps. Come back here for depth.

Page	What it covers	Reach for it when
installation.md	Install on Linux / macOS / Windows / WSL; user vs project scope	Setting wgm up the first time
playbook.md	The per-build SOP, a per-gate PASS/FAIL cheat sheet, reading a docs-audit report	You want the checklist form
running-the-loop.md	loop.sh + the swarm (parallel worktrees), limits, retry/circuit-breaker, the metrics ledger, thresholds, model escalation	Driving a large, parallel, or autonomous build
containers.md	Podman / OCI validation environment	A scenario needs the app actually running
devcontainers.md	Sandboxing the loop itself, disk-consciously	Running an autonomous loop you want isolated
troubleshooting.md	Symptom, cause, and resolution by stage	Something doesn't work

Looking up an exact flag? That lives in the reference — loop.sh, swarm.sh, installers, gates, and artifacts.

Just took delivery of a build you didn't write? See companion skills — /teach-me maps it, /quiz-me checks what stuck, and /rugged stress-tests whether it (or its plan) actually holds up.

See also

• Documentation index — the full map, including the agent/ docs.
• SKILL.md — the authoritative protocol the agent follows.
• Quickstart README — install and invoke wgm.