Openclaw Skill Gastown

WarnAudited by ClawScan on May 10, 2026.

Overview

The skill is coherent for multi-agent coding, but it gives autonomous background agents broad command and code-merge authority with limited explicit approval boundaries.

Install only if you intentionally want a persistent multi-agent coding system. Run setup manually, pin dependency versions if possible, use isolated branches or worktrees, require explicit approval before merges or broad fixes, and inspect/stop background Gas Town workers when finished.

Findings (8)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

A mistaken or overbroad task could lead to many workspace-changing commands being run without the user reviewing each command first.

Why it was flagged

The skill explicitly assigns the agent direct command execution authority for the orchestrator, rather than limiting it to user-reviewed instructions.

Skill content
You operate the system directly - users never run terminal commands themselves. You execute all `gt` and `bd` commands via Bash
Recommendation

Use this only in trusted repositories or disposable worktrees, and require explicit approval before spawning workers, running broad fixes, or merging results.

What this means

Agents or services may keep operating on queued work after the initial request, consuming resources or changing repository state.

Why it was flagged

The architecture includes background daemon behavior and periodic nudging, showing activity that can continue beyond a single user prompt.

Skill content
Deacon ⚙️ Infrastructure daemon. - Background patrol loop - Health checks, session monitoring - Nudges agents periodically
Recommendation

Make start/stop controls explicit, verify what processes are running, and shut down Gas Town services when the task is complete.

What this means

A bad instruction or flawed worker change could propagate from one branch into the main codebase.

Why it was flagged

Parallel worker outputs can flow through an automated merge queue into the main branch; the shown artifacts do not specify a human approval gate for that review.

Skill content
Refinery 🦡 Merge queue processor. - Processes polecat branches from merge queue - Merges to main after review
Recommendation

Require human code review or pull-request approval before any merge to main, and run the system on feature branches by default.

What this means

The skill may consume the user's Claude Code account quota and operate with the local CLI's workspace access.

Why it was flagged

The skill relies on local Claude Code processes, which likely use whatever Claude CLI account/session the user has configured.

Skill content
Each polecat is independent Claude Code process
Recommendation

Confirm which Claude account is logged in, monitor usage, and consider a dedicated account or workspace for large agent swarms.

What this means

A future upstream change could alter behavior from what was reviewed here.

Why it was flagged

The setup script installs remote dependencies using `@latest`, so the installed code can change over time.

Skill content
go install github.com/steveyegge/gastown/cmd/gt@latest ... CGO_ENABLED=0 go install github.com/steveyegge/beads/cmd/bd@latest
Recommendation

Pin reviewed versions of `gt` and `bd`, and inspect upstream repositories before running setup.

What this means

Running that command carelessly could remove or replace system-level Go files.

Why it was flagged

The documentation contains a privileged destructive setup command. It is not shown as automatic, but it can delete an existing system Go installation.

Skill content
sudo rm -rf /usr/local/go && sudo tar -C /usr/local -xzf go1.24.12.linux-amd64.tar.gz
Recommendation

Do not run privileged setup commands automatically; verify paths and prefer the provided user-local setup approach when possible.

What this means

Stale or incorrect persisted instructions could influence later agents or tasks.

Why it was flagged

The skill deliberately persists work state and task context so future agents can resume from it.

Skill content
Git-backed persistent state (hooks) - Work tracking (beads/convoys)
Recommendation

Inspect and clean persisted hooks, Beads data, and work queues between unrelated projects or after sensitive tasks.

What this means

Mistaken or spoofed local agent messages could trigger work transitions or completion signals.

Why it was flagged

Workflow control depends on messages between agents, but the shown artifacts do not describe identity checks or message-origin boundaries.

Skill content
Mail System 📬 Inter-agent communication ... Special messages: SWARM_START, SWARM_COMPLETE, POLECAT_DONE, MERGED
Recommendation

Use the mail system only within trusted local workspaces and review routes, mailboxes, and attached work before letting agents act on them.