Ralph Loop (Agent Mode)
PassAudited by ClawScan on May 10, 2026.
Overview
This instruction-only skill openly runs external coding agents that can edit and commit code, so it appears purpose-aligned but should be used only in a controlled project or sandbox.
Install only if you want your OpenClaw agent to orchestrate external coding CLIs. Run it in a known project directory, preferably in a sandbox or separate git branch, keep max iterations and timeouts limited, avoid yolo or permission-skipping flags unless you deliberately accept the risk, and review the persistent prompt/agent/plan files and resulting commits.
Findings (3)
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.
A coding agent can run for up to the configured timeout, modify files, run tests or other project commands, and potentially commit changes.
The skill directs the agent to launch and monitor external coding CLIs through raw exec/process tools. This is central to the skill’s purpose, but it gives the workflow meaningful authority over the local project.
Use exec tool to launch the coding agent ... background: true ... pty: true ... Use process tool to monitor
Use a sandbox or disposable branch, confirm the working directory and maximum iterations, avoid auto-approval flags unless you understand the risk, and review logs and commits before trusting results.
If PROMPT.md, AGENTS.md, specs, or IMPLEMENTATION_PLAN.md contain incorrect or unsafe instructions, future loop iterations may follow them.
The skill intentionally reuses project-local instruction and planning files across iterations, so stale or unwanted instructions in those files could affect later agent actions.
The loop persists context via `PROMPT.md` + `AGENTS.md` (loaded each iteration) and the plan/specs on disk.
Review these persistent files before and during a loop, especially after an agent updates AGENTS.md or IMPLEMENTATION_PLAN.md.
Users may have less clarity about the exact source lineage of the skill.
This metadata differs from the registry entry for ralph-loop-agent version 1.1.0 and from the package repository information, creating a provenance ambiguity even though no executable code is included.
"owner": "jordyvandomselaar", "slug": "ralph-loop", "latest": { "version": "0.1.1"Verify the publisher and repository history before relying on it in important projects.