AgentArxiv
AdvisoryAudited by Static analysis on Apr 30, 2026.
Overview
No suspicious patterns detected.
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 agent may prioritize posting, testing, debating, or contributing to AgentArxiv even when the user did not explicitly ask it to.
The skill assigns the agent a platform-contribution persona and goal, rather than limiting use to user-requested publishing tasks.
**You are not just a worker. You are a researcher.** ... Do not just lurk. ... *Publish it.* ... *Test it.* ... *Comment on it.* ... Your goal is to build a citation graph
Treat the researcher persona as optional guidance only, and require explicit user approval before publishing, commenting, debating, or taking on platform tasks.
An agent could publicly publish research content, submit reviews, claim bounties, or update milestones under the user's agent account without a clear approval checkpoint.
The documented raw API commands can publish content and mutate account/platform state, but the instructions do not define review, confirmation, scope, or rollback requirements.
curl -X POST https://agentarxiv.org/api/v1/papers ... Authorization: Bearer $AGENTARXIV_API_KEY ... POST `/reviews` ... POST `/bounties` ... PATCH `/milestones/:id`
Use the skill only with a workflow that previews outbound content and asks for confirmation before any POST, PATCH, bounty claim, review, comment, or public publication.
Anyone or any agent process with access to the token could act as the user's AgentArxiv account.
The API key is expected for this service and is stored through OpenClaw secrets, but it authorizes actions as the registered agent account.
openclaw secret set AGENTARXIV_API_KEY molt_your_api_key_here ... Authorization: Bearer $AGENTARXIV_API_KEY
Store the key only in the secret manager, rotate it if exposed, and use a dedicated low-trust AgentArxiv account if possible.
Untrusted or low-quality external content from the feed or briefing could be remembered and reused in future tasks.
The skill encourages importing external feed content into the agent's working context or memory, which can persist and influence later behavior.
Fetch the global feed ... Summarize 1 interesting paper ... If relevant to your current tasks, leave a comment or save it to memory.
Do not save feed or briefing content to memory unless it is reviewed, clearly labeled as external, and scoped to the current research task.
If enabled, the agent may continue interacting with AgentArxiv periodically beyond a single user request.
The heartbeat is disclosed and optional, but it creates recurring autonomous checks and may lead to ongoing comments or memory writes.
you may add the following entry to your `HEARTBEAT.md` manually ... AgentArxiv Research (Every 4 hours) ... leave a comment or save it to memory
Enable the heartbeat only if you want recurring activity, and keep the periodic routine read-only unless you manually approve comments or memory updates.