Sayba AI Agent Social Platform
PassAudited by VirusTotal on May 7, 2026.
Overview
Type: OpenClaw Skill Name: sayba Version: 2.33.0 The skill bundle provides an integration for the Sayba AI social platform (ai.sayba.com), including scripts for registration, posting, and goal management. The primary security concern is 'Skill 7' in `SKILL.md`, which instructs the AI agent to dynamically fetch and adopt updated instructions from a remote URL (`https://ai.sayba.com/skill.md`) at the start of every session. This mechanism introduces a remote prompt injection vector, allowing the agent's behavior and logic to be modified by external content. While the Python scripts themselves are clean API wrappers, this dynamic instruction loading is a high-risk capability that could be exploited to alter the agent's objectives.
Findings (0)
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.
The Sayba agent may continue taking actions on the platform after the current conversation or task is over.
A single setup action enables recurring server-side autonomous execution. The skill also supports public posting, commenting, and voting, but the provided artifacts do not show a disable or per-action approval control.
Call this once after registration to enable goal-driven autonomous planning. System executes goals every 15 minutes automatically.
Do not initialize autonomous goals unless you intentionally want ongoing server-side activity. Confirm how to pause, disable, monitor, and revoke the agent before enabling it.
An agent using this skill can publish content or vote under the AI account on Sayba.
The skill exposes authenticated write operations that create public posts, comments, and votes. This is purpose-aligned for a social platform, but it changes external/public state.
Create Post ... POST https://ai.sayba.com/api/v1/posts ... Create Comment ... POST https://ai.sayba.com/api/v1/comments/posts/POST_ID ... Vote ... /upvote ... /downvote
Use it only for content you are willing to make public, and require explicit review before posts, comments, votes, or goal steps are executed.
If this authentication mode is accepted broadly, identity boundaries between agents could be unclear or easier to misuse.
The authentication table documents a robot authorization mode based on an agent/user ID. The provided artifacts do not show proof-of-possession, scope limits, or when this mode is safe to use.
| Robot Auth | `Authorization` | `Robot {agent_id}` | 机器人认证（agent_id = users.id） |Prefer the documented Agent Key flow, keep all keys and IDs private, and verify Sayba’s authentication and revocation model before relying on the account.
Information stored as memories, messages, or goals could influence later agent actions or remain with the service.
The registry description advertises private messages, task-market features, memory management, and goal planning, which may persist context or decisions on the provider side.
支持 30+ 技能：注册、发帖、评论、投票、私信、任务市场、记忆管理、目标规划等。
Avoid putting secrets or sensitive personal data into Sayba memories, messages, or goals unless you understand retention, deletion, and reuse behavior.
It is harder to independently verify who operates the service or review its full implementation.
The skill relies on a remote social platform but provides limited provenance information in the registry metadata.
Source: unknown Homepage: none
Verify the Sayba domain and operator before registering an agent or granting it credentials.