ClawHub

Resilient Claude Agent

v1.0.17

Run Claude Code sessions in tmux for fire-and-forget execution with crash recovery, model routing, and structured task state.

⭐ 0· 81·0 current·0 all-time
bymatt martinez@mattmartinez
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
VirusTotalVirusTotal
Pending
View report →
OpenClawOpenClaw
Benign
high confidence
✓
Purpose & Capability
Name/description (resilient Claude sessions in tmux) matches the included scripts and declared binary requirements (tmux, claude, openclaw). The files implement lifecycle, monitoring, resume, and atomic manifest updates that you would expect for this purpose.
ℹ
Instruction Scope
The SKILL.md and scripts explicitly instruct running claude with --dangerously-skip-permissions, dispatching commands into tmux via send-keys, writing files in the project dir and temp dirs, and killing tmux sessions that match a prefix. These actions are dangerous but are openly documented and integral to the skill's design; they do broaden the runtime scope (arbitrary shell execution via the orchestrator and manipulation of tmux sessions).
✓
Install Mechanism
Instruction-only with included scripts (no download/extract/install spec). Nothing writes arbitrary remote code at install time; all code is provided in the skill bundle. This is low-risk from an install mechanism perspective.
✓
Credentials
No credentials or sensitive environment variables are required. Only MONITOR_* tuning env vars are referenced optionally. The skill does rely on external binaries (claude, openclaw) which the SKILL.md correctly asks users to audit, but it does not request unrelated secrets or broad system credentials.
✓
Persistence & Privilege
always:false and no special platform privileges requested. The skill creates tmux sessions and temp directories and updates its own manifest files, but it does not claim to persist as an always-enabled skill or modify other skills' configurations.
Assessment
This skill is internally consistent but explicitly enables a powerful, dangerous mode: it runs Claude with --dangerously-skip-permissions and dispatches shell commands into tmux. Only install it if you trust the orchestrator and are willing to run the agent in an isolated account/container/VM. Before installing: (1) audit the claude and openclaw binaries on PATH, (2) avoid using the 'claude-' tmux prefix for other unrelated sessions, (3) prefer a dedicated user or container for these tasks, and (4) consider whether you need an alternative that requires explicit human consent rather than skipping permissions. If you cannot guarantee the orchestrator's trustworthiness or prefer stricter controls, do not install this skill.

Like a lobster shell, security has layers — review code before you run it.

latestvk9711s3ejqbdyadycsvw8d3c29848rnf

License

MIT-0
Free to use, modify, and redistribute. No attribution required.
Termshttps://spdx.org/licenses/MIT-0.html

Runtime requirements

🛡️ Clawdis
Binstmux, claude, openclaw

Comments