agentleFS
doivamongdoivamong/agent-workbench/.claude/skills/awb-install-and-verify/SKILL.md

awb-install-and-verify

Skill2 starsChanged 6 months agoMITView on GitHub

WHAT: drive the kit's own installer in plain language — wire the safety hooks into a project, then run the real --doctor verifier and explain, honestly, what now protects the user and what it does NOT. USE WHEN: a user wants to set up or check the workbench's guards ("install the workbench", "set up the hooks", "are my guards actually on?", "did the install work?", "is the dangerous-command block working?"). DO NOT TRIGGER: building a new feature/hook (that's a plan-then-code skill); editing settings.json by hand (this skill only ever calls install.py); a general "is my code good?" review (that's awb-review).

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 1 is distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Does: run install.py --merge-settings to wire the hooks, run install.py --doctor to verify them, and translate the verifier's output into …

    line 28
  2. Very specific:

    Wire the hooks: Run python install.py <project-path> --merge-settings, where <project-path> is the adopter project (an external path).

    line 37
  3. Very specific:

    Verify — run the real doctor: Run python install.py <project-path> --doctor (here .

    line 41
  4. Very specific:

    HARD GATE: doctor must exit 0: If it prints any FAIL, the guards are not protecting the user yet — relay the doctor's own Fix: line in …

    line 47
  5. Distinctive:

    Do not tell the user they're protected.

    line 49
  6. Very specific:

    If the user asks "is anything exposed?

    line 65

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

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 doctor. Run python install.py <project-path> --doctor (here . is allowed, to verify the kit's own wiring). Read-only; it launches the wired guards and reports. Inside an adopter project (the kit was installed there, you're not in the kit folder), prefer python tools/doctor.py — the same verifier, copied in with the kit, so the user can re-check "are my guards still on?" from their own repo. Fall back to install.py <project-path> --doctor from the kit folder if tools/ wasn't installed.
  3. HARD GATE: doctor must exit 0. If it prints any FAIL, the guards are not protecting the user yet — relay the doctor's own Fix: line in plain language and stop. Do not tell the user they're protected. A green-looking summary you didn't read is not proof.
  4. Relay the result honestly — read the doctor output, assert nothing beyond it.
    • PROVEN [name] → that guard was actually exercised and worked. Today only block_dangerous.py is PROVEN: it really blocked a dangerous command. Say plainly it "catches common destructive commands, but is not a security boundary — a determined command can still get past it."
    • INSTALLED [name] → wired and its interpreter runs, but its behaviour was not tested. Say "set up and ready, but I haven't watched it do its job" — never upgrade an INSTALLED to "working".
  5. Tell them to restart. The doctor proves the scripts run on this machine, not that the running session loaded them. Tell the user to restart Claude Code (or start a new session) so the hooks take effect.

Recovery (owner-scoped, minimal)

If the user asks "is anything exposed? / undo the phone/LAN exposure / am I locked out?": the exposable surface (ops/*, ui/web/, set_password.py) is not in the installer's COPY_MAP, so in a project you installed the kit into, it isn't there at all. Answer plainly: "those features aren't installed in your project, so there's nothing of theirs to expose." Only route to tools that --doctor/the manifest show are actually present; if a step needs a tool that isn't installed, explain the manual step rather than calling a file that doesn't exist.

Anti-rationalization

You'll thinkReality
"The summary said all passed, I'll just say you're protected"Read the per-hook lines. Most are INSTALLED, not PROVEN — saying "protected" oversells what was tested.
"Close enough to call this hook working"INSTALLED ≠ working. The honest word is "wired"; only --doctor's PROVEN earns "working".
"I'll skip the restart note"Without a restart the user's live session may not have the hooks at all — the guards look on but aren't.

See also

The plain-language, user-facing walkthrough — install, uninstall (safety model), and a troubleshooting guide — is the single source in docs/getting-started.md. Point the user there for the full prose; this skill is the agent's playbook for driving the installer.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.