Back to skill

Security audit

warapp-mcp

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious because a scanner found it can accept a surrender that ends a game without asking the user.

Review the surrender-acceptance behavior before installing, especially whether you are comfortable with the skill ending a game without asking. The result is suspicious, not a claim that the author is malicious; Skillspector's coverage is partial because it could not resolve local references.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 222)May include surrounding context.

md
a free-for-all.
  When it is not safe the first warning says so, and whether `warapp_submit_turn` would refuse those
  orders (it does in a Commanders game; an `enemy_deploy` what-if alone does not make it refuse) unless
  `accept_commander_risk: true` — never pass it without asking the user, since losing the commander
  loses the game. Put `needs` armies on the commander's territory instead (or move it away; under
  Local Deployment only a pool that reaches it can be deployed there). The submit counts Airlifts
  (an Airlift off the commander's territory, or of the commander next to the enemy, is refused like

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

This exception authorizes accepting another player's surrender without user approval whenever a tool reports winsGame=true, potentially ending the game on the user's behalf during an otherwise read-only status check. A favorable outcome does not establish authorization, and the skill explicitly acknowledges that its confirmation mechanism may be advisory rather than a human-enforced gate; the game-only scope limits the impact.

Content

Scanner excerpt · SKILL.md (reported line 1085)May include surrounding context.

md
the owner says so — in a free-for-all it is a decision (strategy note `ffa-surrender-veto`): the preview's
  `advice` names the neighbour that gains most from the land, with `land.borderedBy`, `landBecomes` (neutral or
  AI), `othersStillToAccept` and `winsGame`; present them. The one exception: when `winsGame` is `true` (accepting
  simply wins the game) accept without asking, then tell the owner it was done. It refuses when no surrender is
  pending from that player or the user's acceptance is not owed. The result is `accepted` (the seat is out),
  `pending` (the user's acceptance is recorded; others still owe theirs), `unconfirmed` (check `warapp_my_games`)
  or `unknown` (it may have gone through: check, never resend). Chat stays read-only: never reply in chat.