Back to skill

Security audit

Shipyard launch

Security checks for vulnerabilities and agentic risk

Overview

This launch helper needs review because its search-setup workflow can redeploy production apps beyond the one you selected.

Before installing, review the required full plugin and restrict access to the intended app, project and domain. Require approval before redeploying any additional project. Verify that an existing account is a dedicated demo account before resetting its credentials, and use synthetic content for email testing. The clean static scan does not resolve these documented scope concerns.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
Findings (15)

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 23)May include surrounding context.

md
## Rules (always)

- You do the work. Never hand the student a command you could run.
- Never ask for keys in the chat and never read `~/.shipyard/secrets.env` or `.env*` files. Keys go in the keys file via `node "${CLAUDE_PLUGIN_ROOT}/scripts/secrets-init.mjs"`; scripts show only names with a tick or a cross. Live Stripe keys, Resend keys and SMTP passwords never go in argv, files in the repo, or logs.
- Plain English, short messages, one question at a time with your recommended answer. Explain any new word in five words or fewer.
- **Never fake legal certainty.** Legal text and country notes are drafts and research, not legal or tax advice. Say "draft, have it checked" in chat, in `HANDOFF.md` and in the checklist. The public pages themselves do not carry that wording; the admin Health page does.
- Real money only after the student says so, in chat, for that step. Test mode first, always.

Activation keywords include ambiguous standalone terms such as "launch" and "terms" without requiring an app-launch context.

Medium
Category
Trigger Abuse
Confidence
93% confidence
Finding

The description explicitly activates the skill for isolated words such as "launch" and "terms", which can appear in unrelated requests. Although the description explains the skill's app-launch purpose, it does not require that context for these triggers or exclude unrelated uses.

Content

No source excerpt is available for this finding.

The email-testing step sends an email to a third-party service without a privacy warning.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

This workflow instructs the student to send an email to mail-tester.com, rather than merely listing the service. It does not warn that the service will receive the message contents and headers, which may contain personal information or other sensitive data.

Content

No source excerpt is available for this finding.

The workflow instructs the agent to redeploy an app without warning the user about production changes.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The Search Console discovery workflow directs the agent to redeploy a project when its domain returns NOT_FOUND, potentially including projects discovered outside the originally selected app. Redeployment can publish unintended code or affect production availability, and this step does not require confirming project scope, deployment readiness, or authorization before making that change.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
75% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · references/stores-audit.md (reported line 10)May include surrounding context.

md
this.
- **OAuth in the system browser, back by deep link.** Google blocks its sign-in inside embedded web views. Open OAuth in the system browser (`@capacitor/browser`), return to a custom scheme or universal link (for example `myapp://auth/callback`), catch it with `@capacitor/app` `appUrlOpen`, and exchange the code on a native callback route. Add the deep-link address to the Supabase auth redirect URL list.
- **Permission strings.** Every native permission used needs a clear reason in `Info.plist` (`NSMicrophoneUsageDescription`, `NSCameraUsageDescription`, `NSPhotoLibraryUsageDescription` and so on) saying what the app does with it. Ask only for what the app uses; on Android remove unused permissions from the manifest.
- **Apple 4.2, minimum functionality.** A website in a frame gets rejected. Have at least one native piece in the main flow (share sheet, haptics, push, native status bar and splash, microphone capture), a designed offline screen instead of a browser error, and no l

MCP server referenced without pinned version: 'npx @capacitor/assets'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

Although this is not an MCP server, npx @capacitor/assets can download and execute a registry package without a reviewed version because no local installation of that package is specified. A compromised future release could execute with the build user's privileges and access source code, available credentials, or signing material.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx cap'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx cap'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx cap'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx cap'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

The seed-script instructions reset existing account passwords and mark emails verified without warning about the account-access implications.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The seed instructions explicitly accommodate existing users and then require an administrative password reset and email-confirmation override without verifying that the account is a dedicated demo account. If account lookup selects a real user, this can bypass email ownership verification, grant access through a known demo password, and disrupt the user's existing access.

Content

No source excerpt is available for this finding.

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 · references/stores.md (reported line 11)May include surrounding context.

md
- The app must be live on a custom or at least stable HTTPS address (`go-live.md` steps 1 to 4). The shell loads that address.
- Account deletion in the app and public privacy and terms pages are required by the stores; the template has them.
- Payments: Apple requires its own in-app purchase for digital subscriptions sold inside an iOS app in most regions. Do not put Stripe buttons for digital goods in the iOS app without checking the rules for the student's region. Android has its own rules too; check Google Play's current payments policy.

## Costs and people

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 · references/stores.md (reported line 28)May include surrounding context.

md
- Phone: `node "${CLAUDE_PLUGIN_ROOT}/scripts/add-shell.mjs" <appDir> mobile`. It refuses without a live HTTPS address. It writes a `mobile/` folder with a Capacitor wrapper (a thin app that shows your site), an offline page, icons, the iOS privacy file, `STORE.md` (listing draft) and a `README.md` with the exact next commands. App id comes from `--app-id` or is derived; it must be a reverse domain like `com.example.myapp`. Choose it once: it cannot change after publishing.
- Desktop: `add-shell.mjs <appDir> desktop` writes `desktop/` (Electron) and a GitHub workflow that builds on tags named `desktop-v*`.
- Do not run `npx cap add ios` or `npx cap add android` or open Xcode/Android Studio for them without asking: they need those tools installed. If the student wants to, follow the shell's README with them and say what each command does.

## Check

MCP server referenced without pinned version: 'npx cap'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx cap'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.