Perry Coding Agents

Dispatch coding tasks to OpenCode or Claude Code on Perry workspaces. Use for development work, PR reviews, or any coding task requiring an isolated environment.

This skill appears to do what it says (dispatch remote coding agents), but there are several red flags you should address before installing or using it: - Missing declared requirements: The SKILL.md assumes local tools (tailscale, jq, ssh, curl) and a webhook token but the skill metadata declares none. Confirm these prerequisites and provide them securely. - Undeclared secret: The instructions require an Authorization: Bearer <hooks-token> value. Ask the author to declare required env vars (e.g., PERRY_HOOKS_TOKEN) or explain how tokens are provisioned. Never paste secrets into a skill without understanding storage/use. - Insecure SSH option: The examples use -o StrictHostKeyChecking=no which disables host key verification and increases MITM risk. Prefer adding known host keys or using a secure SSH config rather than disabling checking. - Local HTTP callback risk: Remote workspaces are instructed to POST back to your local IP:18789. Ensure that the wake endpoint is bound only to a safe interface, is authenticated, and is not exposed to untrusted networks. Consider firewall/Tailscale ACLs to limit who can reach that port. - Remote code execution: The skill runs arbitrary commands on remote workspaces via SSH. Only use with workspaces and keys you fully trust. Review remote binaries paths (/home/workspace/.opencode/, /home/workspace/.local/bin/claude) and confirm they are the intended agents. What would reduce my concern: explicit metadata listing required binaries and environment variables (including hook token), instructions that avoid disabling host-key checking, and clear guidance on securing and scoping the local wake endpoint. If the author provides those clarifications, the skill would look coherent and likely benign; until then treat it cautiously.

Type: OpenClaw Skill Name: perry-coding-agents Version: 1.5.0 The skill bundle is classified as suspicious primarily due to the use of `ssh -o StrictHostKeyChecking=no` in `SKILL.md`. While this might be intended for automation in trusted environments, it disables host key verification, making SSH connections vulnerable to Man-in-the-Middle (MITM) attacks. Other aspects, such as the `curl` callback to the agent's own IP for task completion, are aligned with the stated purpose and do not show malicious intent. The instructions for the AI agent are workflow-related and do not constitute malicious prompt injection.