Claude Code Orchestrator (tmux-first)

WarnAudited by ClawScan on May 10, 2026.

Overview

The skill matches its stated tmux-orchestration purpose, but it runs background Claude Code sessions with permissions bypassed and broad shell/SSH authority, so it needs careful review before use.

Install only if you intentionally want a background Claude Code runner. Use it in a disposable worktree or clean git branch, verify the workdir and SSH host, avoid secrets in transcripts, monitor the tmux session, and be ready to kill the session or revert changes if the task drifts.

Findings (7)

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 task can run tools and change project files without per-action confirmation, so a wrong prompt, wrong directory, or unexpected Claude Code behavior could cause broad changes.

Why it was flagged

The normal workflow explicitly bypasses Claude Code permission prompts while delegating a development task to another agent.

Skill content
Launch `claude --dangerously-skip-permissions` in interactive mode.
Recommendation

Use only in a clean git branch or sandboxed worktree, require explicit user approval for this mode, and prefer normal Claude Code permissions or a narrower allow-list where possible.

What this means

A path containing shell metacharacters or an incorrectly constructed workdir could cause unintended commands to run in the tmux session.

Why it was flagged

The user-supplied workdir is inserted into a shell command without shell-quoting before the command is submitted in tmux.

Skill content
tmux -S "$SOCKET" send-keys -t "$SESSION":0.0 -l -- "cd $WORKDIR && export https_proxy=http://127.0.0.1:6152 http_proxy=http://127.0.0.1:6152 all_proxy=socks5://127.0.0.1:6153 && claude --dangerously-skip-permissions"
Recommendation

Validate workdir values and shell-quote them, for example by using `cd -- "$WORKDIR"` in a generated script or `printf %q`/`shlex.quote` before sending commands to tmux.

What this means

A Claude Code task may continue consuming resources or modifying files after the user stops watching unless the session is explicitly checked or killed.

Why it was flagged

The skill starts detached tmux sessions that can keep running outside the immediate chat turn.

Skill content
tmux -S "$SOCKET" new -d -s "$SESSION" -n shell
Recommendation

Add a clear timeout, cleanup command, and explicit user confirmation before starting or restarting long-running sessions.

What this means

If pointed at a sensitive SSH alias or production host, the skill can exercise that account's remote privileges for coding tasks.

Why it was flagged

Remote mode uses the user's SSH/scp access to copy files and run tmux/Claude Code on another host.

Skill content
scp -q "$PROMPT_FILE" "${SSH_HOST}:${REMOTE_REF}"
Recommendation

Use a dedicated, least-privilege SSH account or development host, and confirm the target alias and workdir before running remote tasks.

What this means

Recent task output may include code, file paths, errors, or secrets and could be forwarded for summarization.

Why it was flagged

The documented summary workflow can send task JSON to the OpenClaw gateway; the same section states entries include `lastLines`, which are recent tmux transcript lines.

Skill content
bash {baseDir}/scripts/list-tasks.sh --json | \
  openclaw gateway call summarize-tasks --stdin
Recommendation

Review transcript content before sending summaries, keep captured line counts low, and avoid running this on sessions that may display secrets.

What this means

Task names and delivery history can persist across runs and may reveal project context or affect future reconciliation output.

Why it was flagged

The package includes persistent delivery-state files with task labels and timestamps, and the reconciler updates similar state.

Skill content
"tasks": {
    "ccmerge-next-steps": {
      "deliveredReportMtime": 1771079894,
Recommendation

Clear packaged state on install if it is not needed, and document what state is stored, where, and how to reset it.

What this means

Users may not realize what local tools and trust assumptions are required until execution time.

Why it was flagged

The registry metadata provides no upstream source/homepage and does not declare runtime binaries, even though the included scripts invoke multiple local tools.

Skill content
Source: unknown
Homepage: none
Required binaries (all must exist): none
Recommendation

Declare required binaries and provenance in metadata, and verify the bundled scripts before use.