SaySigned - Agreement infrastructure for AI agents

Enable AI agents to register, create, send, sign, and verify legally binding e-signature agreements via API or email with automated key and config management.

This skill appears to be a real e-signature integration, but it deserves careful handling. Before installing or enabling it: (1) Verify the service domain (saysigned.com) and company legitimacy independently. (2) Prefer to perform registration and paste the API key yourself rather than letting the agent write it into global MCP config; if you must store a key, create a limited-scope key and rotate it regularly. (3) Disable autonomous signing unless you explicitly want the agent to sign agreements on your behalf — require per-envelope human approval for any signature. (4) Do not allow the agent to restart system services or modify configs without an explicit confirmation dialog that the user sees and approves. (5) Be cautious about webhooks/URLs you provide (they may receive sensitive document events). If you want a safer setup, keep the key out of global configs and require manual confirmation for every sign/send action.

Type: OpenClaw Skill Name: saysigned Version: 1.0.1 The OpenClaw AgentSkills skill bundle for SaySigned appears benign. The `SKILL.md` documentation is comprehensive, clearly outlining the purpose and usage of the e-signature service. Crucially, it includes explicit instructions for the AI agent to require human approval for sensitive operations like modifying MCP client configuration, storing API keys, restarting the MCP client, and initiating billing upgrades (by showing the `checkout_url`). This strong emphasis on human-in-the-loop for critical actions significantly mitigates prompt injection risks. There is no evidence of intentional malicious behavior, data exfiltration to unauthorized endpoints, persistence mechanisms, or obfuscation. The `webhook_url` parameter, while a potential vector for data exfiltration if an attacker could inject a malicious URL, is a standard feature and the skill itself does not instruct its misuse; it's a vulnerability risk if the agent is compromised, not malicious intent from the skill developer.