Back to skill

Security audit

Job Agent for AgentMesh360

Security checks across malware telemetry and agentic risk

Overview

This skill is not clearly malicious, but it can automatically submit real job applications and messages using saved job-site sessions without a final approval step.

Install only if you are comfortable with an agent using your job-platform logins and API key to submit selected applications/messages automatically after preview. Review the selected list carefully before starting a round, use a dedicated environment or browser profile if possible, and be aware that the installer and updates rely on a remote GitHub repository.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (7)

Vague Triggers

Medium
Confidence
90% confidence
Finding
The manifest description advertises broad resume-driven job discovery and automatic delivery across multiple job platforms without clear trigger constraints, user consent boundaries, or a limited activation scope. That increases the chance the skill is invoked in contexts where the user does not expect real external-account actions, especially because later instructions explicitly automate application delivery.

Missing User Warnings

High
Confidence
98% confidence
Finding
The skill directs the agent to automatically send real job applications on external platforms after review, while framing the preview as 'required notice, not another confirmation' and repeatedly instructing the agent not to ask again. This is dangerous because it enables consequential real-world actions affecting the user's reputation, job search, and third-party accounts without a strong upfront consent checkpoint immediately before execution.

Natural-Language Policy Violations

Medium
Confidence
80% confidence
Finding
Forcing a Chinese-language message without checking the user's preferred locale can cause the user to misunderstand account-credit status and the fact that the workflow will continue into live actions. In a skill that can trigger external submissions, degraded comprehension materially increases the risk of uninformed consent and accidental actions.

Autonomous Decision Making

Medium
Category
Excessive Agency
Content
- Never invent an API Key. Ask the user to create an AgentMesh360 universal Key at `https://agentmesh360.com/app/` and wait. Registration and Key creation are free; cloud capabilities require available credits.
- After configuring the Key, run `jobagent doctor env`. Read `environment_healthy` and `workflow.ready` separately. If `cloud_access.usable=true`, briefly report the active balance source and run the top-level `next_suggested` immediately. Never block on `Pass: not purchased`; ask for a purchase only when `paid_pass_required=true` or a real cloud command returns `insufficient_credits`.
- New accounts start with zero cloud credits. For grandfathered `signup_trial_active`, tell the user: `你的 AgentMesh360 账户仍有此前发放的体验额度:剩余 {credit} credits,有效期至 {expires_at}。无需购买通行证,我现在继续执行下一步。` Then execute `next_suggested` without asking for confirmation.
- Run Boss直聘 -> 猎聘 -> 智联招聘 -> 51Job as complete vertical chains. Never pre-login future platforms; complete the current platform's `login -> discover -> review -> send -> audit` chain and complete its audit before logging in to the next platform. Never operate their shared browser concurrently.
- Stop whenever `requires_user_action=true`; relay `user_prompt` exactly and wait.
- Report `selected / review / rejected`, then automatically deliver the signed `selected` list without asking again.
Confidence
96% confidence
Finding
without asking

Autonomous Decision Making

Medium
Category
Excessive Agency
Content
- New accounts start with zero cloud credits. For grandfathered `signup_trial_active`, tell the user: `你的 AgentMesh360 账户仍有此前发放的体验额度:剩余 {credit} credits,有效期至 {expires_at}。无需购买通行证,我现在继续执行下一步。` Then execute `next_suggested` without asking for confirmation.
- Run Boss直聘 -> 猎聘 -> 智联招聘 -> 51Job as complete vertical chains. Never pre-login future platforms; complete the current platform's `login -> discover -> review -> send -> audit` chain and complete its audit before logging in to the next platform. Never operate their shared browser concurrently.
- Stop whenever `requires_user_action=true`; relay `user_prompt` exactly and wait.
- Report `selected / review / rejected`, then automatically deliver the signed `selected` list without asking again.
- On `event=delivery_preview`, show every row in `delivery_preview.items` before any real platform action. Prefer a compact table using the declared columns; otherwise relay `delivery_preview.fallback_text` unchanged. This is required notice, not another confirmation. Then execute the exact top-level `next_suggested`, including `--preview-id`.
- On `error=delivery_preview_required`, immediately run the returned review `next_suggested`, display the regenerated preview, and follow its newly bound send command. Preserve existing user promotions; do not recollect, recharge or ask for confirmation.
- Show `skipped_delivered` when present and never add those jobs back to the send list.
Confidence
99% confidence
Finding
without asking

Autonomous Decision Making

Medium
Category
Excessive Agency
Content
- After an existing installation updates, run `jobagent upgrade-check` and resolve its `next_suggested` action before opening a platform. Never delete `~/.jobagent` or the Job Agent Chrome profile as a general fix; preserve credentials, login cookies, profiles, audits and preferences.
- Forward `client_update_detected -> client_update_started -> client_update_completed -> client_command_resumed` once in the user's language. Do not ask permission for a managed signed update and do not stop after success; continue the original command. Stop only on `client_update_failed`, report its `message`, and follow `next_suggested`. Older clients may first emit only the compatibility completion/resume pair.
- For `client_update_failed` with `error_code=release_artifact_hash_mismatch`, run the returned official-installer recovery command once and repeat the original command. It preserves Job Agent state and browser sessions; never disable the signature/tag/commit/archive checks or delete the managed profile.
- When a cloud command returns `retryable=true` and `request_preserved=true`, do not ask the user to retry, re-login or recollect jobs. Run the returned `next_suggested` command immediately; Discover resumes the same signed request and candidate set without another charge.
- Profiles, rounds, decisions and audits are account-bound. On `local_state_owner_required`, ask the user to confirm ownership and run `jobagent account bind --confirm-legacy`. On `local_state_account_mismatch`, ask the user to confirm switching accounts and run `jobagent account switch --new-state`. Never edit account-state files manually.
- Diagnose browser slowness or conflicting login evidence with `jobagent browser diagnose --platform <platform>` before asking for another login. Treat `login.state=unknown` or `conflicting` as inconclusive.
- Forward CLI progress stages and heartbeats. Use compact `jobagent round audit` by default; expand only failures or explicitly requested details.
Confidence
87% confidence
Finding
do not ask the user

Session Persistence

Medium
Category
Rogue Agent
Content
## Safety Contract

- Never invent an API Key. Ask the user to create an AgentMesh360 universal Key at `https://agentmesh360.com/app/` and wait. Registration and Key creation are free; cloud capabilities require available credits.
- After configuring the Key, run `jobagent doctor env`. Read `environment_healthy` and `workflow.ready` separately. If `cloud_access.usable=true`, briefly report the active balance source and run the top-level `next_suggested` immediately. Never block on `Pass: not purchased`; ask for a purchase only when `paid_pass_required=true` or a real cloud command returns `insufficient_credits`.
- New accounts start with zero cloud credits. For grandfathered `signup_trial_active`, tell the user: `你的 AgentMesh360 账户仍有此前发放的体验额度:剩余 {credit} credits,有效期至 {expires_at}。无需购买通行证,我现在继续执行下一步。` Then execute `next_suggested` without asking for confirmation.
- Run Boss直聘 -> 猎聘 -> 智联招聘 -> 51Job as complete vertical chains. Never pre-login future platforms; complete the current platform's `login -> discover -> review -> send -> audit` chain and complete its audit before logging in to the next platform. Never operate their shared browser concurrently.
Confidence
72% confidence
Finding
create an AgentMesh360 universal Key at `https://agentmesh360.com/app/` and wait. Registration and Key creation are free; cloud capabilities require available credits. - After configuring the Key, run

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.