claude-code (Deprecated alias)

SuspiciousAudited by ClawScan on May 10, 2026.

Overview

This is a coherent Claude Code tmux orchestrator, but it deserves review because it bypasses Claude Code permissions and has unsafe shell argument handling.

Install/use this only if you intentionally want Claude Code to run development work without per-command permission prompts. Run it in a trusted repo, clean branch, or sandbox; verify workdir values are safe; avoid secrets in terminal output; use trusted SSH hosts only; and monitor or stop tmux sessions when done.

Findings (6)

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 coding task can run commands and edit files without per-action Claude Code approval; a mistaken prompt, unsafe reference file, or wrong workdir could cause unwanted changes.

Why it was flagged

The script starts Claude Code inside tmux with permission checks disabled and then submits the task automatically, giving the spawned agent broad ability to act in the user's environment.

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

Use only in a sandbox, clean git branch, or disposable workspace. Prefer normal Claude Code permissions where possible, and require explicit user approval before launching this mode.

ConcernHigh Confidence
ASI05: Unexpected Code Execution
What this means

If the workdir value is derived from untrusted text or contains characters like semicolons or command substitutions, unintended commands may run as the user.

Why it was flagged

The workdir argument is inserted into a shell command without shell escaping before it is sent to tmux. A path containing shell metacharacters could execute extra commands.

Skill content
--workdir) WORKDIR="$2"; shift 2 ;; ... "cd $WORKDIR && export https_proxy=... && claude --dangerously-skip-permissions"
Recommendation

Quote or shell-escape workdir, label, session, socket, and remote command arguments; validate paths before use; or invoke a wrapper script that receives arguments safely.

NoteHigh Confidence
ASI10: Rogue Agents
What this means

Claude Code work may continue after the initial chat turn and may consume resources or keep changing files until the session finishes or is stopped.

Why it was flagged

The skill intentionally creates long-running tmux sessions and callback-driven task delivery. This is disclosed and purpose-aligned, but it is persistent automation.

Skill content
Start a dedicated tmux session. ... Always include callback via wrapper ... If no pane output for >2-3 min, inspect and restart session. Kill stale Claude processes before restart.
Recommendation

Monitor sessions with the provided attach/status commands, stop stale sessions deliberately, and avoid overlapping labels that could kill or replace the wrong session.

What this means

Using remote mode gives the skill the same access your SSH key/account has on that host and copies selected task materials there.

Why it was flagged

Optional remote mode uses the user's SSH identity to copy prompt files and run tmux/Claude Code commands on a remote host.

Skill content
scp -q "$PROMPT_FILE" "${SSH_HOST}:${REMOTE_REF}" ... ssh -o BatchMode=yes "$SSH_HOST" "tmux -S '$SOCKET' new -d -s '$SESSION' -n shell"
Recommendation

Use only trusted SSH aliases and hosts, prefer limited-purpose keys/accounts, and confirm what prompt/report files will be copied.

What this means

Terminal output, build logs, or task text could be included in summaries; those logs may contain sensitive project details or secrets if commands print them.

Why it was flagged

The documented summary flow can include recent tmux pane output and send it through an OpenClaw gateway summarization call.

Skill content
Each entry contains: ... lastLines ... bash {baseDir}/scripts/list-tasks.sh --json | openclaw gateway call summarize-tasks --stdin
Recommendation

Avoid running summaries on sessions that may display secrets, reduce captured lines, and review/redact reports or transcripts before sharing.

What this means

The skill may fail or use whatever matching binaries are present on PATH, which can surprise users in mixed or untrusted environments.

Why it was flagged

The script checks for runtime dependencies, while the registry metadata declares no required binaries. This is a dependency-contract gap rather than evidence of malicious behavior.

Skill content
for tool in tmux claude rg python3 git; do
Recommendation

Declare required tools in metadata and verify local PATH/tool provenance before use.