OpenCode ACP Control v2

Control OpenCode directly via the Agent Client Protocol (ACP). Start sessions, send prompts, resume conversations, and manage OpenCode updates. Includes auto...

This skill appears to implement what it promises (starting and managing a local OpenCode ACP session), but there are a few red flags you should check before installing: - The metadata declares no required binaries, but the script and docs require an 'opencode' binary and the 'jq' tool; ensure those are present and trusted on your system. - The included script will delete lock files under ~/.openclaw/agents and may kill processes; this is intended for cleanup but can affect other agents or sessions on the same host—backup important state and understand the command scope before running. - There is no homepage or verified source; the author/email in SKILL.md and a GitHub URL are present in docs but the registry 'source' is unknown. Prefer installing only from a verified repository or after manual inspection of the script. - If you plan to integrate with external MCPs (Supabase/GitHub), the skill mentions them but does not request credentials—you will need to supply and protect any required tokens separately. If you want higher confidence: verify the opencode binary you will use (origin, checksums), ensure jq is installed from a trusted package manager, and manually review the opencode-session.sh contents (especially the find/rm commands) to confirm the cleanup logic matches your expectations.

Type: OpenClaw Skill Name: opencode-acp-control-v2 Version: 2.2.1 The skill bundle provides a framework for controlling the OpenCode Agent Client Protocol (ACP), which requires high-risk capabilities including background process execution and shell-based file manipulation. Specifically, SKILL.md and opencode-session.sh instruct the agent to execute shell commands like 'rm -f' and 'find' to manage lock files in the ~/.openclaw directory. While these actions are plausibly necessary for the stated purpose of session recovery and management, the reliance on broad shell execution and the instruction for the AI to manage its own process lifecycle constitute a significant attack surface without robust input sanitization.