Forge

Create, build, review, repair, and validate complete Agent Skill packages through a six-phase pipeline, outputting finished installable skill files.

Forge is coherent for its purpose: it will build and write complete, installable skill packages and keep logs/journals under ~/openclaw. Before installing or enabling it, consider: (1) Require manual review/approval of any package Forge produces before installation — do not auto-install outputs. (2) Restrict or monitor its filesystem and network capabilities if your environment allows policy controls (prevent writing to unexpected locations or pulling arbitrary GitHub code). (3) Vet the upstream GitHub source the skill references if you allow its self-update feature. (4) Monitor journals/decisions files it writes and intake processing to ensure it only processes expected Mentor proposals. These controls will reduce risk from legitimately powerful but high-impact behavior. If you want lower risk, run Forge in an isolated/test environment and inspect build outputs before trusting them in production.

Type: OpenClaw Skill Name: ocas-forge Version: 2.3.0 The 'ocas-forge' skill implements a self-update mechanism in SKILL.md (forge.update) that uses the GitHub API to download a tarball and overwrite its own source directory using shell commands (tar, cp, rm). While this is functionally consistent with a development tool, it creates a high-risk remote code execution (RCE) vector and supply-chain vulnerability. Furthermore, the forge.init process establishes persistence by automatically registering a daily cron job to execute this update, which could be leveraged to maintain control over the agent if the remote repository is compromised.