awb-install-and-verifylisted
Install: claude install-skill doivamong/agent-workbench
# Install the workbench and prove the guards are on
> **Announce on activation:** "Using awb-install-and-verify — I'll wire the hooks, then run --doctor and tell you exactly what's protected."
The persona this kit serves often can't read a stack trace or a settings file. So setup
must end in a plain-language, *truthful* answer to one question: "which guards are on now,
and which are not?" The only honest source for that answer is the kit's own `--doctor`,
which runs the guards and reports back. This skill makes the agent run it and relay it —
never narrate protection it hasn't proven.
## Scope
- **Does:** run `install.py --merge-settings` to wire the hooks, run `install.py --doctor`
to verify them, and translate the verifier's output into plain language (what's PROVEN,
what's only INSTALLED, what failed and how to fix it).
- **Does NOT:** hand-edit `.claude/settings.json` (only `install.py` touches it); claim a
hook *works* when `--doctor` only proves it is wired; call any tool that isn't installed
in the target project (see Recovery). It is a bypassable exemplar, not a gate.
## Process
1. **Wire the hooks.** Run `python install.py <project-path> --merge-settings`, where
`<project-path>` is the **adopter project** (an external path). (Skip if the user only wants
to verify an existing install.) Note: `install.py` refuses to install into the kit repo
itself, so do **not** use `.` here — only `--doctor` (step 2) accepts `.`.
2. **Verify — run the real doct